Imaging Pearls ❯ March 2025
|
-- OR -- |
|
3D and Workflow
- Throughout my career, I have learned that anyone can be wrong anytime. Yet, through these experiences, we stumbled upon a pivotal insight, one that enabled us to stand out in a competitive market: the power of open source allows our work to transcend personal gain and contribute to a collective higher purpose. It is important to note that as leaders we have a responsibility for every single person, and we have a responsibility to live with integrity, to be impeccable with our word, and to live the ideals that we are proposing. If we achieve this, it becomes not about us but about the mission.
Life: Distributed and Open Source
Mullenweg M, Fishman EK, Chu LC, Rowe SP, Rizk RC
J Am Coll Radiol. 2024 Aug 21:S1546-1440(24)00696-3. doi: 10.1016/j.jacr.2024.08.005. Epub ahead of print. PMID: 39155028. - “Furthermore, as the leader, especially of a large organization, it is important to be able to move between the 40,000- foot and 1-inch perspectives. Attention to detail, at every level, is essential for great leaders.”
Life: Distributed and Open Source
Mullenweg M, Fishman EK, Chu LC, Rowe SP, Rizk RC
J Am Coll Radiol. 2024 Aug 21:S1546-1440(24)00696-3. doi: 10.1016/j.jacr.2024.08.005. Epub ahead of print. PMID: 39155028. - In preparing for the future of technology, I think it is good to take a barbell approach. For example, I like to read things that are either very new, such as the latest papers, along with material that has survived a long time, such as the works of Aristotle or Plato. In between this time frame, you find that where the truth and the real is, and that is where you should focus.
Life: Distributed and Open Source
Mullenweg M, Fishman EK, Chu LC, Rowe SP, Rizk RC
J Am Coll Radiol. 2024 Aug 21:S1546-1440(24)00696-3. doi: 10.1016/j.jacr.2024.08.005. Epub ahead of print. PMID: 39155028. - “Because radiology is one of the most rapidly evolving fields in medicine, it is critical we as radiologists think ahead. Adopting the barbell approach within radiology positions the field to stay at the forefront of medical science by balancing high-level, visionary thinking with a focus on past works that have survived for long periods of time. For instance, although embracing new technologies such as artificial intelligence and machine learning for image analysis, the barbell approach would allow us to appreciate Albert Einstein’s warning, “We should take care not to make the intellect our god; it has, of course, powerful muscles, but no personality”
Life: Distributed and Open Source
Mullenweg M, Fishman EK, Chu LC, Rowe SP, Rizk RC
J Am Coll Radiol. 2024 Aug 21:S1546-1440(24)00696-3. doi: 10.1016/j.jacr.2024.08.005. Epub ahead of print. PMID: 39155028. - “As patient care is the priority of our field, this quote serves as a reminder that behind every image and diagnosis are human beings with fears, hopes, and needs. It gives us the perspective to balance technological and scientific advancements with compassion and patient-centered care. The focus on lessons from the past and progress in the present enables the field to adopt the latest advancements as well as refine and perfect them, making certain that the field remains effective in its mission to provide the best possible care for our patients. We need to make sure we keep this in our focus.”
Life: Distributed and Open Source
Mullenweg M, Fishman EK, Chu LC, Rowe SP, Rizk RC
J Am Coll Radiol. 2024 Aug 21:S1546-1440(24)00696-3. doi: 10.1016/j.jacr.2024.08.005. Epub ahead of print. PMID: 39155028.
Adrenal
- “They are mainly represented by adrenal cavernous hemangiomas (ACHs) and adrenal cystic lymphangiomas (ACLs). Both ACHs and ACLs are usually unilateral, benign, non-functioning, and asymptomatic, but differential diagnosis with other adrenal masses before surgical removal may be challenging. Symptoms, such as dull abdominal or flank pain, may occur if tumor size increases, thus determining mass effect on adjacent structures. In addition, the risk of life-threatening retroperitoneal hemorrhage in proportionally related to tumor size. Radiological characteristics often overlap with pheochromocytomas and adrenal carcinomas, therefore, ruling out malignancy becomes mandatory.”
The diagnostic dilemma of adrenal vascular tumors: analysis of 21 cases and systematic review of the literature.
Coscia K, Ravaioli C, Tucci L, et al.
Endocrine. 2025 Jan 18. doi: 10.1007/s12020-024-04123-5. Epub ahead of print. PMID: 39825193. - “ACHs and ACLs are benign vascular neoplasms consisting of many entangled thin-walled and aberrant dilated vessels that are prone to rupture. The first case of adrenal cavernous hemangioma (ACH) was reported in 1955 by Johnson and Jeppesen, while adrenal cystic lymphangioma (ACL) was first described in 1965 by Linn. ACHs and ACLs usually arise from the adrenal cortex, as only two cases of adrenal medulla involvement have been described. All of our cases originated in the adrenal cortex, with only one case associated with medullary hyperplasia. Dilated spaces delimitated by a single endothelial layer are typical histopathological features of ACHs and ACLs. ACHs are usually associated with necrotic and hemorrhagic areas separated by fibrotic septa.”
The diagnostic dilemma of adrenal vascular tumors: analysis of 21 cases and systematic review of the literature.
Coscia K, Ravaioli C, Tucci L, et al.
Endocrine. 2025 Jan 18. doi: 10.1007/s12020-024-04123-5. Epub ahead of print. - “ACHs and ACLs have been detected with increased frequency due to the improvement of imaging techniques. However, they represent a diagnostic dilemma in clinical practice due to their rarity and their misleading imaging features overlapping with adrenal malignant tumors. Because of the heterogeneous clinical and radiological pictures, treatment should be targeted to the patient’s characteristics. Therefore, if an adrenal vascular tumor is suspected, treatment options must be discussed by a multidisciplinary team including endocrinologists, radiologists, pathologists, and surgeons, with expertise in the differential diagnosis of adrenal tumors.”
The diagnostic dilemma of adrenal vascular tumors: analysis of 21 cases and systematic review of the literature.
Coscia K, Ravaioli C, Tucci L, et al.
Endocrine. 2025 Jan 18. doi: 10.1007/s12020-024-04123-5. Epub ahead of print.
Cardiac
Coronary Artery Disease Reporting and Data System Update
Arzu Canan, Fernando Uliana Kay
Rad Clinics NA 2025 (in press)- The CAD-RADS category represents the highest grade of stenosis (reduction in vessel diameter) detected on coronary CTA, taking into account the assessment of all sizable coronary arteries (>1.5 mm). There are 7 categories, detailed in Table 1. CAD-RADS 0 indicates the absence of atherosclerotic plaque or stenosis, while CADRADS 5 signifies total coronary artery occlusion. The categories between 0 and 5 describe varying degrees of stenosis:.
CAD-RADS 0: No plaque, no stenosis
CAD-RADS 1: <25% stenosis
CAD-RADS 2: 25%-49% stenosis
CAD-RADS 3: 50% to 69% stenosis - CAD-RADS 4: 70% to 99% stenosis CAD-RADS 4 is further divided into 2 subgroups based on the number of vessels involved and the presence of 50% left main stenosis:.
CAD-RADS 4A: Severe stenosis in 1 or 2 vessels
CAD-RADS 4B: Severe stenosis in 3 vessels or 50% left main stenosis
CAD-RADS 5: Total occlusion
Coronary Artery Disease Reporting and Data System Update
Arzu Canan, Fernando Uliana Kay
Rad Clinics NA 2025 (in press) - Further Recommendations and Management Takes account of plaque burden and ischemia assessment
CAD-RADS 0 No further testing or management
CAD-RADS 1–2 No further testing; preventive measures P1/P2: risk factor modification and preventive pharmacotherapy P3/P4: Aggressive risk factor modification and preventive pharmacotherapy
CAD-RADS 3 Functional assessment and, Aggressive risk factor modification and preventive pharmacotherapy for all plaque groups If modifier I1, consider ICA
CAD-RADS 4–5 Functional assessment or ICA and, Aggressive risk factor modification and preventive pharmacotherapy for all plaque groups Other treatments per guideline
Coronary Artery Disease Reporting and Data System Update
Arzu Canan, Fernando Uliana Kay
Rad Clinics NA 2025 (in press)
Chest
- Upper limits of normal aortic diameter by segment: aortic root 4.0 cm, ascending aorta 4.0 cm or less than 1.5 times the descending aortic diameter, aortic arch 3.5 cm, descending aorta 3.0 cm. The new classification for aortic dissection from 2020 describes the entry tear zone, which determines the dissection type A/B, followed by the subscripts that denote the proximal and distal extensions according to the involved aortic zones. Although penetrating aortic ulcer (PAU) is categorized as part of the acute aortic syndrome , the vast majority of PAUs are asymptomatic. Symptomatic PAU is often associated with aortic wall hemorrhage. Ascending aortic dilation ranges from 4.0 to 4.4 cm, with an aneurysm defined at 4.5 cm.
Preoperative Imaging of the Thoracic Aorta
Zehavit E. Kirshenboim et al.
Radiol Clin N Am - (2024) in press - Aortic diameters vary by age and gender. For simplicity, the following upper limits of normal can be used: aortic root 4.0 cm, ascending aorta 4.0 cm or less than 1.5 times the descending aortic diameter, aortic arch 3.5 cm, descending aorta 3.0 cm. The normal ascending aortic wall thickness is approximately 2 mm.
Preoperative Imaging of the Thoracic Aorta
Zehavit E. Kirshenboim et al.
Radiol Clin N Am - (2024) in press - The aorta dilates with aging at a rate of approximately 0.7 mm per decade in women and 0.9 mm per decade in men. Thickening of the aortic wall and degeneration of the collagen and elastic components lead to aortic dilation, elongation, and tortuosity with advancing age, a process known as arteriosclerosis. Repeated pulsatile stress causes fragmentation of the elastic components in the proximal aorta, replacement by fibrotic tissue, and resultant aortic wall stiffening and increased pulse pressure, increasing the left ventricular workload.
Preoperative Imaging of the Thoracic Aorta
Zehavit E. Kirshenboim et al.
Radiol Clin N Am - (2024) in press - Acute IMH is contained hemorrhage within the aortic media, which can be attributed to PAU, trauma, or thrombosed dissected false lumen with microscopic tears in the intima. IMH accounts for 5% to 15% of AAS cases and occurs more frequently in older patients (60–80 years of age) compared to those with aortic dissection. Other risk factors include hypertension and aortic dilation, with commonly found concomitant abdominal aortic aneurysm. IMH is classified similarly to aortic dissection, with most cases (58%) being a type B.
Preoperative Imaging of the Thoracic Aorta
Zehavit E. Kirshenboim et al.
Radiol Clin N Am - (2024) in press - The two most common forms of inflammatory (noninfectious) aortitis are Takayasu arteritis and giant cell arteritis (GCA). Takayasu (necrotizing) arteritis typically affects females under 40 year old and usually manifests as panaortitis with granulomatous inflammation which may cause stenosis of the inflamed vessels, particularly the arch branches. GCA is more common in patients over 50 year old, is associated with temporal arteritis and polymyalgia rheumatica, syphilis (known as luetic aortopathy), mycobacterium tuberculosis infection, and human immunodeficiency virus. Contrast-enhanced CT is typically the preferred imaging modality for diagnosis, with key radiological findings including aortic wall thickening, periaortic fluid or soft tissue, rapid development of saccular aneurysms, and occasionally, the presence of air within the aortic wall. The term “mycotic aneurysm” refers to aneurysms caused by an infection, which are characterized by a mushroom-shaped appearance, and does not indicates fungal infection.
Preoperative Imaging of the Thoracic
Aorta Zehavit E. Kirshenboim et al.
Radiol Clin N Am - (2024) in press
- “The majority of intrathoracic goiters are substernal extensions of cervical goiters. They commonly occur in the superior portion of the mediastinum and may extend to the anterior and middle mediastinum. Approximately 2% of cases show no evident connection to the cervical thyroid gland and are thought to arise from an ectopic thyroid gland in the mediastinum. It might be difficult to differentiate preoperatively intrathoracic goiters arisen from an ectopic thyroid from other anterior mediastinal solid tumors such as thymic epithelial tumors. Adenomatous goiter is the most common histologically, and carcinomas occur in 2–38% of intrathoracic goiters. Thyroid tissues may show high attenuation due to intrinsic iodine on plain CT . Adenomatous goiters show multiple nodules, and calcifications, cystic change , and hemorrhage in the lesion are common.”
Anterior mediastinal lesions: CT and MRI features and differential diagnosis
Takahiko Nakazono · Ken Yamaguchi · Ryoko Egashira1 ·et al.
Japanese Journal of Radiology (2021) 39:101–117
- Indications for Thymic Imaging
• Incidental finding in the prevascular (anterior) mediastinum
• Chest symptoms: chest pain, dysphagia, dyspnea, persistent cough, hoarseness
• Paraneoplastic syndromes: myasthenia gravis, Lambert-Eaton syndrome, pure red cell aplasia, hypogammaglobulinemia, lichen planus, Good syndrome, limbic encephalitis
• Staging and follow-up after resection of a thymic epithelial neoplasm - “One of the most common pitfalls in evaluation of a thymic lesion at CT is misinterpreting a simple thymic cyst as a solid lesion. Thymic cysts may contain proteinaceous or hemorrhagic fluid . This results in attenuation measurements greater than those expected for simple fluid (>20 HU), which are reported to be up to 97 HU. In a large series, Ackman et al (30) reported a mean attenuation value of 25 HU for individual cysts with more than 5 years of follow-up . Therefore, when a homogeneously smooth well-defined saccular, oval, or round lesion is detected in the thymic area, even if it measures higher than 20 HU, it is advisable to conduct further evaluation using MRI.”
Thymic Imaging Pitfalls and Strategies for Optimized Diagnosis.
Klug M, Strange CD, Truong MT, et al.
Radiographics. 2024 May;44(5):e230091. - “In rare cases, thymomas may be present as predominantly necrotic or cystic simulating thymic cysts. Another common pitfall with CT is failing to recognize the features of a complex cyst, including intracystic nodularity, a thick capsule, and thick septa. Thus, when the CT features of a cystic prevascular mass are equivocal and a complex cyst is suspected, MRI should be performed.”
Thymic Imaging Pitfalls and Strategies for Optimized Diagnosis.
Klug M, Strange CD, Truong MT, et al.
Radiographics. 2024 May;44(5):e230091. - “Another pitfall encountered with CT is misinterpretation of thymic hyperplasia as thymic malignancy. At CT, the presence of a diffusely enlarged thymus, exhibiting nodularity and a pyramidal configuration, with increased length and thickness of each lobe relative to age-specific measurements, should raise suspicion for thymic hyperplasia. This phenomenon is often seen in patients after undergoing a stressor such as chemotherapy, radiation therapy, or steroid treatment; after recovery from burns or injuries; or in conjunction with systemic diseases such as myasthenia gravis or collagen vascular disease. The presence of intralesional fat intercalating between thymic tissue is diagnostic of thymic hyperplasia.”
Thymic Imaging Pitfalls and Strategies for Optimized Diagnosis.
Klug M, Strange CD, Truong MT, et al.
Radiographics. 2024 May;44(5):e230091. - “There are some typical interpretation pitfalls and blind spots for metastatic involvement in the assessment of patients with thymic malignancies. Thymic malignancies have a predilection for pleural involvement . Pleural thickening, nodules, or masses are the most common radiologic signs of pleural dissemination; pleural effusion is rare. Pleural metastases can be lenticular shaped and relatively flat and may be missed when they are small. Thin-section CT is superior to thick-section CT for evaluation of pleural metastases. The thinner 1.25-mm sections, with the improved spatial resolution of the lung kernel, can help to identify early pleural nodularity in comparison with the usually thicker 2.5-mm sections used with the soft-tissue kernel. Moreover, coronal images may be helpful for detecting subtle diaphragmatic surface pleural metastases.”
Thymic Imaging Pitfalls and Strategies for Optimized Diagnosis.
Klug M, Strange CD, Truong MT, et al.
Radiographics. 2024 May;44(5):e230091. - “Thymic carcinoma, constituting 14–22% of TET, demonstrates a lower incidence compared to thymomas. Patients with thymic carcinoma often manifest symptoms related to mediastinal mass lesions, with high frequencies of invasion into surrounding organs, lymph node metastases, and distant metastases, resulting in a poor prognosis. Paraneoplastic syndromes that are commonly present in patients with thymoma are very rare. Imaging findings typically show irregular margins, presenting as irregular or lobulated forms, with cystic degeneration, hemorrhage, and necrosis observed inside the tumor. The distinction from thymic carcinoma and high-risk thymomas on imaging can be challenging, but thymic carcinoma tends to exhibit internal heterogeneity and higher rates of infiltration into the surrounding structures, along with increased frequency of distant metastases.”
Imaging of thymic epithelial tumors-a clinical practice review.
Koyasu S.
Mediastinum. 2024 Jun 7;8:41. doi: 10.21037/med-23-66. PMID: 39161582; PMCID: PMC11330907. - “NET arising from the thymus, characterized by the dominance or near-total presence of neuroendocrine cells in TET, account for 2–5% of TET. Most occur in adult patients. All thymic neuroendocrine neoplasms (NENs), which includes both NET and neuroendocrine carcinomas, share a propensity for recurrence, lymph node or distant metastasis, and tumor-associated death, with increasing risk from low-grade to high-grade tumors. Their radiological appearance is basically indistinguishable from that of thymic carcinomas. These tumors, classified into atypical carcinoids and typical carcinoids, often exhibit large, irregularly margined masses without distinct capsules on imaging.”
Imaging of thymic epithelial tumors-a clinical practice review.
Koyasu S.
Mediastinum. 2024 Jun 7;8:41. doi: 10.21037/med-23-66. - “Thymic cysts are different from TET (thymomas, thymic carcinomas, and thymic NETs as above), which are solid neoplasms that may show heterogeneous enhancement, necrosis, invasion, or calcification. However, thymic cysts are often misinterpreted as solid lesions such as thymic epithelial neoplasms only by CT, or sometimes even by MRI because thymic cysts often showed features suggestive of intralesional microbleeding, inflammation, and fibrosis. Recent study describes that most thymic cysts changed in volume [31 of 34 cysts (91%)], CT attenuation [15 of 35 cysts (43%)], and T1-weighted MRI signal [12 of 18 cysts (67%)] over more than 5 years of follow-up, although none developed mural irregularity, nodularity, or septations.
Imaging of thymic epithelial tumors-a clinical practice review.
Koyasu S.
Mediastinum. 2024 Jun 7;8:41. doi: 10.21037/med-23-66. - T-LBLs have a predilection for rapid dissemination and the tumor spreads to the extrathoracic lymph nodes, bone marrow, and central nervous system in extensive disease. In malignant lymphomas, penetration of vessels may be seen, and calcification is very rare prior to chemotherapy. Dynamic contrast-enhanced MRI has been reported to show gradual enhancement in malignant lymphomas. The ADC values of malignant lymphomas are reported to be very low, reflecting the high cellularity in the tumor.
Anterior mediastinal lesions: CT and MRI features and differential diagnosis
Takahiko Nakazono · Ken Yamaguchi · Ryoko Egashira1 ·et al.
Japanese Journal of Radiology (2021) 39:101–117 - Thymolipoma is a rare benign tumor that contains fat and non-neoplastic thymic tissues. The average age of patients is 22–26 years, with no gender predominance, and most patients are asymptomatic. Thymolipomas typically show a large, well-defined, and soft mass in the anterior mediastinum and pericardial region and may mimic cardiomegaly on chest radiograph. CT and MRI show a mass comprised of intermingled soft tissue and fat tissue in the anterior mediastinum. Differential diagnosis of thymolipoma includes lipoma and liposarcoma.
Anterior mediastinal lesions: CT and MRI features and differential diagnosis
Takahiko Nakazono · Ken Yamaguchi · Ryoko Egashira1 ·et al.
Japanese Journal of Radiology (2021) 39:101–117
- “The average long-axis dimension was 17.50 ± 6.00 mm, the CT value range across the 24 cases was 5–81 HU, and the average CT value of the noncontrast enhanced scans was 39.75 ± 20.66 HU. The CT value in the noncontrast enhanced scan was >20 HU in 79% of the sample cases.”
Special Computed Tomography Imaging Features of Thymic Cyst.
He ZL, Wang ZY, Ji ZY.
Int J Clin Pract. 2022 Oct 11;2022:6837774 - “Thymic cysts are considered to be a relatively rare type of anterior mediastinal mass; they are reported to account for approximately 1%–3% of cases. With the growing popularity of low-dose chest computed tomography (CT) screening, the detection rate of thymic cysts is also increasing. The density of thymic cysts is often not the typical liquid density, and the CT diagnosis rate is not high, leading to thymic cysts often being misdiagnosed as thymomas or other types of solid mass and being removed by resection.”
Special Computed Tomography Imaging Features of Thymic Cyst.
He ZL, Wang ZY, Ji ZY.
Int J Clin Pract. 2022 Oct 11;2022:6837774 - “The thymic cysts in the present study were all located in the anterior mediastinum. The average long-axis dimension measurement was 17.50 ± 6.00 mm. All 24 cases had uniform density. The average noncontrast enhanced scan CT value was 39.75 ± 20.66 HU, and the lowest and highest noncontrast enhanced scan CT values were approximately 5 and 81 HU, respectively. Among the sample, 19/24 cases (79%) had a noncontrast enhanced scan CT value of > 20 HU.”
Special Computed Tomography Imaging Features of Thymic Cyst.
He ZL, Wang ZY, Ji ZY.
Int J Clin Pract. 2022 Oct 11;2022:6837774 - “Most of the existing literature reports that thymic cysts are typically triangular or round in shape, with smooth edges, rare lobes, and smooth junctions with the pleura. Some thymic cysts can display changes in shape between different scan phases. Most cases in the present study were triangle/chestnut-shaped (50%) or round (29%), and the majority of the cysts had smooth edges.”
Special Computed Tomography Imaging Features of Thymic Cyst.
He ZL, Wang ZY, Ji ZY.
Int J Clin Pract. 2022 Oct 11;2022:6837774 - Due to the issue of atypical density, the main concern in the differential diagnosis of thymic cysts is thymoma. Larger thymomas can more easily be distinguished from thymic cysts due to their uneven internal density, progression, and invasion of the surrounding structures, combined with more obvious enhancement. However, the density of smaller thymomas is relatively uniform, and there exist significant image overlaps between thymic cysts and some mildly enhanced tumors. This is also the reason for most cases of misdiagnosis of thymic cysts as thymoma
Special Computed Tomography Imaging Features of Thymic Cyst.
He ZL, Wang ZY, Ji ZY.
Int J Clin Pract. 2022 Oct 11;2022:6837774 - “Mature teratomas are benign mediastinal germ cell tumors, typically containing fat, fluid, calcification, and soft tissue . In a series of 66 mediastinal mature teratomas evaluated with CT, soft tissue was observed in 100%, fluid in 88%, fat in 76%, and calcifications in 53%. Mature teratomas are more common in young patients, accounting for approximately 25% of anterior mediastinal masses in ages 10 to 19 and less than 5% over CT, mature teratomas characteristically have a well-defined, lobulated, smooth margin with a heterogenous appearance.”
Multimodality imaging of mediastinal masses and mimics.
Archer JM, Ahuja J, Strange CD, et al.
Mediastinum. 2023 May 8;7:27. doi: 10.21037/med-22-53. - “The parathyroid glands are typically located along the posterior border of the thyroid gland. Parathyroid glands located above or below the thyroid gland in the neck or mediastinum are considered ectopic. The inferior parathyroid glands are more commonly ectopic, located in the mediastinum in approximately 4% to 5% of the population. Patients can present with incidentally detected hypercalcemia. Symptoms of hypercalcemia can include muscle pain, lethargy, nausea, constipation, and confusion.”
Multimodality imaging of mediastinal masses and mimics.
Archer JM, Ahuja J, Strange CD, et al.
Mediastinum. 2023 May 8;7:27. doi: 10.21037/med-22-53. - “Although the majority of parathyroid adenomas occur in isolation, several genetic syndromes have been associated with parathyroid neoplasia, including multiple endocrine neoplasia (MEN) types 1 and 2A. MEN 1 and 2A are autosomal dominant conditions . Hyperparathyroidism in these conditions is typically multiglandular. The classic components of MEN 1 include parathyroid tumors, pancreatic islet cell tumors, and pituitary tumors. Additional associations in MEN 1 include facial angiofibromas, adrenal cortical tumors, lipomas, and carcinoid tumors. MEN 2A is characterized by medullary thyroid carcinoma, pheochromocytomas, and parathyroid hyperplasia or tumors.”
Multimodality imaging of mediastinal masses and mimics.
Archer JM, Ahuja J, Strange CD, et al.
Mediastinum. 2023 May 8;7:27. doi: 10.21037/med-22-53. - Tumors arising in the thymus are of various histological types, but three types of TET are the most frequent: thymomas, thymic carcinomas, and thymic neuroendocrine tumors (NETs). Regardless of these subtypes, the initial step in the diagnostic process involves distinguishing between solid and cystic lesions. This is crucial as anterior mediastinal masses, including TET, often require differentiation from conditions such as pericardial cysts and thymic cysts, which are common cystic lesions in the anterior mediastinum. Contrast-enhanced CT proves invaluable in this context, with a post-contrast enhancement of generally 20 Hounsfield units (HU) or more indicative of a solid lesion. Conversely, 10–15 HU change in attenuation can be due to various non-specific factors, such as incorrect placement of the region of interest, patient motion, or beam hardening from adjacent enhancing structures. There may be high rate of unnecessary thymectomy due to misinterpretation of thymic cysts, thymic hyperplasia, and lymphoma as thymoma on chest CT.
Imaging of thymic epithelial tumors-a clinical practice review.
Koyasu S.
Mediastinum. 2024 Jun 7;8:41. doi: 10.21037/med-23-66. PMID: 39161582; PMCID: PMC11330907. - “Thymoma, the most common anterior mediastinal tumor, predominantly manifests in middle-aged individuals (range, 55–65 years) and is infrequent in children. Gender predilection is not observed. Often asymptomatic, it is frequently incidentally discovered through imaging studies. Thymoma is associated with various autoimmune disorders, with severe myasthenia gravis occurring in 17–54% of cases as mentioned above. The prevalences of myasthenia gravis are also reported to vary by ethnicity, with some reports ranging from 5.7–82.4%.”
Imaging of thymic epithelial tumors-a clinical practice review.
Koyasu S.
Mediastinum. 2024 Jun 7;8:41. doi: 10.21037/med-23-66. PMID: 39161582; PMCID: PMC11330907. - Primary mediastinal malignant lymphomas commonly occur in the anterior mediastinum. Major histologic subtypes are primary mediastinal large B-cell lymphoma (PMBCL), nodular sclerosis Hodgkin lymphoma (NSHL), and T-cell lymphoblastic lymphoma (T-LBL) . PMBCLs occur in younger adults with a peak incidence at 20–30 years PMBCLs show a large irregular or lobulated mass without a peripheral capsule . The lesions commonly invade adjacent mediastinal structures such as vessels, chest wall, and lung .Necrosis, cystic change and hemorrhage in the lesion, and mediastinal lymph node enlargement are common . Pleural and pericardial effusion are seen in about half of cases.
Anterior mediastinal lesions: CT and MRI features and differential diagnosis
Takahiko Nakazono · Ken Yamaguchi · Ryoko Egashira1 ·et al.
Japanese Journal of Radiology (2021) 39:101–117 - “NSHLs are common in young females. NSHLs typically show a lobulated or multinodular mass with limited necrosis and cystic change. Fibrous septa may be seen in the mass, and the peripheral capsule is unclear. Lymph node enlargement may be seen in regions adjacent to the primary lesion. Pleural and pericardial effusion are not frequent. T-LBLs occur in children and young adults with a male predilection. T-LBL patients may complain of chest pain and dyspnea due to rapid enlargement of the mediastinal lesion T-LBLs show a large heterogenous mass with necrosis. Pleural and pericardial effusion are seen in about half of cases. T-LBLs have a predilection for rapid dissemination and the tumor spreads to the extrathoracic lymph nodes, bone marrow, and central nervous system in extensive disease. In malignant lymphomas, penetration of vessels may be seen, and calcification is very rare prior to chemotherapy. “
Anterior mediastinal lesions: CT and MRI features and differential diagnosis
Takahiko Nakazono · Ken Yamaguchi · Ryoko Egashira1 ·et al.
Japanese Journal of Radiology (2021) 39:101–117 - HL is the most common lymphoma presenting with mediastinal lymphadenopathy and most frequently involves lymph nodes in anterior mediastinal and paratracheal areas in a contiguous manner, and thus involves in decreasing order of frequency the nodes in the hilar, subcarinal, peridiaphragmatic, paraesophageal, and internal mammary areas . Nodular sclerosing HL, the commonest subtype, has a unique predilection for the nodes in the anterior mediastinum.
Cross-Sectional Evaluation of Thoracic Lymphoma
Young A Bae, Kyung Soo Lee
Thorac Surg Clin 20 (2010) 175–186 - “On CT, HL is characterized by the presence of a discrete anterior mediastinal mass with a lobulated contour. The tumor most commonly demonstrates homogeneous soft-tissue attenuation, although large lymph node masses may demonstrate heterogeneity with complex low attenuation representing necrosis, hemorrhage, or cystic degeneration. In the series by Hopper and colleagues, necrotic and cystic-appearing mediastinal lymph nodes were noticed at presentation in 21% of cases of HL. Necrosis is observed most commonly in the nodular sclerosing and mixed cellularity cell types of HL and was not seen in the lymphocyte predominant variety.”
Cross-Sectional Evaluation of Thoracic Lymphoma
Young A Bae, Kyung Soo Lee
Thorac Surg Clin 20 (2010) 175–186 - “Although there are many subtypes of NHL, large B-cell lymphoma and lymphoblastic lymphoma are the most common subtypes, primarily involving the anterior mediastinum. Primary mediastinal large B-cell lymphomas usually present with large and lobulated anterior mediastinal masses and occur predominantly in young adults with a median age of 26 years . Low attenuation areas of necrosis within the mass were seen in 50% and calcification in 5%.13 Also they often directly invade adjacent structures. Lymphoblastic lymphomas are highly aggressive and high-grade lymphomas, arising from thymic lymphocytes.”
Cross-Sectional Evaluation of Thoracic Lymphoma
Young A Bae, Kyung Soo Lee
Thorac Surg Clin 20 (2010) 175–186 - “Dystrophic calcification may develop in involved lymph nodes following mediastinal radiation. The time interval between radiation and the appearance of calcification may be 1 to 9 years. Lymph node calcification before treatment is unusual, but has been associated with aggressive HL or NHL.”
Cross-Sectional Evaluation of Thoracic Lymphoma
Young A Bae, Kyung Soo Lee
Thorac Surg Clin 20 (2010) 175–186 - “Primary pulmonary lymphoma is rare and is encountered usually in NHL. The frequency of lymphoma arising from the lung is estimated to be less than 1% of all lymphomas. The disease usually takes the form of bronchus-associated lymphoid tissue (BALT) lymphoma. In the BALT lymphoma, tumor infiltration develops in multiple extranodal mucosal sites through the lungs. According to a report,38 BALT lymphoma may manifest diverse patterns of lung abnormality on CT, but single or multiple nodule (or nodules) and area (or areas) of consolidation are the main patterns that occur in a majority.”
Cross-Sectional Evaluation of Thoracic Lymphoma
Young A Bae, Kyung Soo Lee
Thorac Surg Clin 20 (2010) 175–186 - “Thoracic lymphomas most frequently involve mediastinal lymph nodes in the anterior mediastinum and paratracheal areas. The lymphomas may also involve lung, thymus, pleura, pericardium, chest wall, and the breast and their radiologic manifestations are diverse. Lymphomas (mostly BALT lymphoma and large B-cell lymphoma) may arise primarily from the lung with various imaging features including single or multiple nodule(s) and area(s) of consolidation. CT is currently the most important imaging modality for the evaluation of thoracic lymphoma but FDGPET also plays a crucial role in the clinical management of these cases.”
Cross-Sectional Evaluation of Thoracic Lymphoma
Young A Bae, Kyung Soo Lee Thorac
Surg Clin 20 (2010) 175–186
- “The risk of aortic dissection and rupture is directly correlated with aortic diameter, with a sharp increase in risk when the ascending aorta measures 6.0 cm or the descending aorta measures 7.0 cm. At this size, there is a 14.1% risk of rupture, dissection, or death per year. For this reason, intervention is recommended when the ascending aorta reaches 5.5 cm or the descending aorta reaches 6.5 cm. For cases of familial aortic aneurysm or connective tissue disease, the threshold is even lower.”
Imaging of the Postoperative Thoracic Aorta
William Truesdell, Smita Pate
Radiol Clin N Am - (2024) in press - “Supracoronary ascending aortic replacement (SCAAR) involves resection of the ascending aorta distal to the sinuses of Valsalva and replacement with an interposition graft. When combined with concurrent replacement of the aortic valve, this is termed the Wheat procedure. This technique can be used to repair the ascending aorta when the underlying pathology spares the aortic annulus and root. SCAAR leaves the native coronary anatomy in place, avoiding potential complications associated with coronary buttons, and has the additional benefit of retaining the flow dynamics of the native coronary ostia. A drawback of the SCAAR is that there may be continued dilation of the aortic root, necessitating reoperation with root replacement.”
Imaging of the Postoperative Thoracic Aorta
William Truesdell, Smita Pate
Radiol Clin N Am - (2024) in press - “One of the most feared complications of endovascular stent grafts is spinal cord ischemia. This can occur from occlusion of the spinal arteries, and is at the greatest risk if the stent graft extends inferiorly below the T8 vertebral body, likely due to coverage of the artery of Adamkiewicz. Interestingly, this complication is less common after treatment of chronic type B aortic dissections, likely due to existing collateral circulation of the spinal cord. This is not directly diagnosed on postoperative CT surveillance; however it can be suggested in symptomatic patients with stent graft coverage of the related spinal arteries.”
Imaging of the Postoperative Thoracic Aorta
William Truesdell, Smita Pate
Radiol Clin N Am - (2024) in press
- “Artificial intelligence (AI) in mammography screening has shown promise in retrospective evaluations, but few prospective studies exist. PRAIM is an observational, multicenter, real-world, noninferiority, implementation study comparing the performance of AI-supported double reading to standard double reading (without AI) among women (50–69 years old) undergoing organized mammography screening at 12 sites in Germany. Radiologists in this study voluntarily chose whether to use the AI system. From July 2021 to February 2023, a total of 463,094 women were screened (260,739 with AI support) by 119 radiologists. Radiologists in the AI-supported screening group achieved a breast cancer detection rate of 6.7 per 1,000, which was 17.6% (95% confidence interval: +5.7%, +30.8%) higher than and statistically superior to the rate (5.7 per 1,000) achieved in the control group.”
Nationwide real-world implementation of AI for cancer detection in population-based mammography screening.
Eisemann N, Bunk S, Mukama T, et al.
Nat Med. 2025 Jan 7. doi: 10.1038/s41591-024-03408-6. Epub ahead of print. - “The recall rate in the AI group was 37.4 per 1,000, which was lower than and noninferior to that (38.3 per 1,000) in the control group (percentage difference: −2.5% (−6.5%, +1.7%)). The positive predictive value (PPV) of recall was 17.9% in the AI group compared to 14.9% in the control group. The PPV of biopsy was 64.5% in the AI group versus 59.2% in the control group. Compared to standard double reading, AI-supported double reading was associated with a higher breast cancer detection rate without negatively affecting the recall rate, strongly indicating that AI can improve mammography screening metrics.”
Nationwide real-world implementation of AI for cancer detection in population-based mammography screening.
Eisemann N, Bunk S, Mukama T, et al.
Nat Med. 2025 Jan 7. doi: 10.1038/s41591-024-03408-6. Epub ahead of print. 
Nationwide real-world implementation of AI for cancer detection in population-based mammography screening.
Eisemann N, Bunk S, Mukama T, et al.
Nat Med. 2025 Jan 7. doi: 10.1038/s41591-024-03408-6. Epub ahead of print.- “To evaluate the potential of AI integration to reduce reading workload through automation, we analyzed a fictitious scenario in which the screening examinations triaged as normal by AI were not read by radiologists. Rather, after an AI prediction of ‘normal’, the examination directly received the final classification ‘normal’, and thus, it would not be possible that any breast cancer signs missed by AI were detected by the radiologists, that a recall was made or that a cancer was detected. The analysis of this scenario showed that, when all normal-tagged examinations (56.7%) were automatically classified as normal, the BCDR was still higher and statistically superior by 16.7% (4.9%, 29.9%), the consensus rate was lower by −19.4% (−21.5%, −17.4%), the recall rate was statistically superior and lower by −15.0% (−18.6%, −11.2%), whereas the biopsy rate was higher by 5.8% (−2.7%, 15.0%) in the AI group than in the control group .”
Nationwide real-world implementation of AI for cancer detection in population-based mammography screening.
Eisemann N, Bunk S, Mukama T, et al.
Nat Med. 2025 Jan 7. doi: 10.1038/s41591-024-03408-6. Epub ahead of print. - “Our study has several strengths. Besides the high number of participants and the real-world setting, likely leading to more conservative effects, a strength of the study is its prospective design. Retrospective analyses are limited by information bias as the final outcomes for examinations identified as suggestive of cancer only by AI are usually unknown. Our study overcomes this limitation as the AI predictions in the study group were considered by radiologists while making clinical decisions. Another strength of PRAIM is the extensive reporting of subgroup results. They showed noninferior or even statistically superior BCDRs in AI-supported screening across screening rounds, breast densities and ages. Thus, AI can be considered for the full screening population and does not need stratified use. Although not a limitation or a strength, it is worth noting that the data used for this study were collected in the early stages of AI use by radiologists (a learning phase). The interaction behavior between radiologists and AI, and hence the screening program metrics achieved, might change as radiologists become more familiar with using the technology.”
- “In conclusion, our findings substantially add to the growing body of evidence suggesting that AI-supported mammography screening is feasible and safe and can reduce workload. Our study also demonstrates that integrating AI into the screening workflow can improve the BCDR with a similar or even lower recall rate. The important downstream effects of AI-supported screening on overall program performance metrics, including interval cancer rate and stage-at-diagnosis distribution at subsequent screening rounds, are subject to follow-up investigations. Nevertheless, based on the now available evidence on breast cancer detection, recall rates, PPV of biopsy and time savings, urgent efforts should be made to integrate AI-supported mammography into screening guidelines and to promote the widespread adoption of AI in mammography screening programs.”
Nationwide real-world implementation of AI for cancer detection in population-based mammography screening.
Eisemann N, Bunk S, Mukama T, et al.
Nat Med. 2025 Jan 7. doi: 10.1038/s41591-024-03408-6. Epub ahead of print.
Deep Learning
- As artificial intelligence continues to shape the medical imaging landscape, its integration into the field of urologic oncology has led to impressive results. For prostate cancer diagnostics, machine learning has shown promise in refining clinically-significant lesion detection, with some success in deciphering ambiguous lesions on multiparametric MRI. For kidney cancer, radiomics has emerged as a valuable tool for better distinguishing between benign and malignant renal masses and predicting tumor behavior from CT or MRI scans. Meanwhile, in the arena of bladder cancer, there is a burgeoning emphasis on prediction of muscle invasive cancer and forecasting disease trajectory. However, many studies showing promise in these areas face challenges due to limited sample sizes and the need for broader external validation.
AI-powered radiomics: revolutionizing detection of urologic malignancies.
Gelikman, David G.a; Rais-Bahrami, et al.
Current Opinion in Urology 34(1):p 1-7, January 2024. - Radiomics integrated with artificial intelligence offers a pioneering approach to urologic oncology, ushering in an era of enhanced diagnostic precision and reduced invasiveness, guiding patient-tailored treatment plans. Researchers must embrace broader, multicentered endeavors to harness the full potential of this field.
AI-powered radiomics: revolutionizing detection of urologic malignancies.
Gelikman, David G.a; Rais-Bahrami, et al.
Current Opinion in Urology 34(1):p 1-7, January 2024.
- “Artificial intelligence (AI) in mammography screening has shown promise in retrospective evaluations, but few prospective studies exist. PRAIM is an observational, multicenter, real-world, noninferiority, implementation study comparing the performance of AI-supported double reading to standard double reading (without AI) among women (50–69 years old) undergoing organized mammography screening at 12 sites in Germany. Radiologists in this study voluntarily chose whether to use the AI system. From July 2021 to February 2023, a total of 463,094 women were screened (260,739 with AI support) by 119 radiologists. Radiologists in the AI-supported screening group achieved a breast cancer detection rate of 6.7 per 1,000, which was 17.6% (95% confidence interval: +5.7%, +30.8%) higher than and statistically superior to the rate (5.7 per 1,000) achieved in the control group.”
Nationwide real-world implementation of AI for cancer detection in population-based mammography screening.
Eisemann N, Bunk S, Mukama T, et al.
Nat Med. 2025 Jan 7. doi: 10.1038/s41591-024-03408-6. Epub ahead of print. - “The recall rate in the AI group was 37.4 per 1,000, which was lower than and noninferior to that (38.3 per 1,000) in the control group (percentage difference: −2.5% (−6.5%, +1.7%)). The positive predictive value (PPV) of recall was 17.9% in the AI group compared to 14.9% in the control group. The PPV of biopsy was 64.5% in the AI group versus 59.2% in the control group. Compared to standard double reading, AI-supported double reading was associated with a higher breast cancer detection rate without negatively affecting the recall rate, strongly indicating that AI can improve mammography screening metrics.”
Nationwide real-world implementation of AI for cancer detection in population-based mammography screening.
Eisemann N, Bunk S, Mukama T, et al.
Nat Med. 2025 Jan 7. doi: 10.1038/s41591-024-03408-6. Epub ahead of print.
Nationwide real-world implementation of AI for cancer detection in population-based mammography screening.
Eisemann N, Bunk S, Mukama T, et al.
Nat Med. 2025 Jan 7. doi: 10.1038/s41591-024-03408-6. Epub ahead of print.- “To evaluate the potential of AI integration to reduce reading workload through automation, we analyzed a fictitious scenario in which the screening examinations triaged as normal by AI were not read by radiologists. Rather, after an AI prediction of ‘normal’, the examination directly received the final classification ‘normal’, and thus, it would not be possible that any breast cancer signs missed by AI were detected by the radiologists, that a recall was made or that a cancer was detected. The analysis of this scenario showed that, when all normal-tagged examinations (56.7%) were automatically classified as normal, the BCDR was still higher and statistically superior by 16.7% (4.9%, 29.9%), the consensus rate was lower by −19.4% (−21.5%, −17.4%), the recall rate was statistically superior and lower by −15.0% (−18.6%, −11.2%), whereas the biopsy rate was higher by 5.8% (−2.7%, 15.0%) in the AI group than in the control group .”
Nationwide real-world implementation of AI for cancer detection in population-based mammography screening.
Eisemann N, Bunk S, Mukama T, et al.
Nat Med. 2025 Jan 7. doi: 10.1038/s41591-024-03408-6. Epub ahead of print. - “Our study has several strengths. Besides the high number of participants and the real-world setting, likely leading to more conservative effects, a strength of the study is its prospective design. Retrospective analyses are limited by information bias as the final outcomes for examinations identified as suggestive of cancer only by AI are usually unknown. Our study overcomes this limitation as the AI predictions in the study group were considered by radiologists while making clinical decisions. Another strength of PRAIM is the extensive reporting of subgroup results. They showed noninferior or even statistically superior BCDRs in AI-supported screening across screening rounds, breast densities and ages. Thus, AI can be considered for the full screening population and does not need stratified use. Although not a limitation or a strength, it is worth noting that the data used for this study were collected in the early stages of AI use by radiologists (a learning phase). The interaction behavior between radiologists and AI, and hence the screening program metrics achieved, might change as radiologists become more familiar with using the technology.”
- “In conclusion, our findings substantially add to the growing body of evidence suggesting that AI-supported mammography screening is feasible and safe and can reduce workload. Our study also demonstrates that integrating AI into the screening workflow can improve the BCDR with a similar or even lower recall rate. The important downstream effects of AI-supported screening on overall program performance metrics, including interval cancer rate and stage-at-diagnosis distribution at subsequent screening rounds, are subject to follow-up investigations. Nevertheless, based on the now available evidence on breast cancer detection, recall rates, PPV of biopsy and time savings, urgent efforts should be made to integrate AI-supported mammography into screening guidelines and to promote the widespread adoption of AI in mammography screening programs.”
Nationwide real-world implementation of AI for cancer detection in population-based mammography screening.
Eisemann N, Bunk S, Mukama T, et al.
Nat Med. 2025 Jan 7. doi: 10.1038/s41591-024-03408-6. Epub ahead of print.
- The application of artificial intelligence (AI) in medicine is expanding at an astonishing pace, mirroring the rapid advances in AI technology itself. Some experts within the field predict that in the next several years, developers may realize artificial general intelligence (AGI)—a revolutionary form of AI capable of understanding, learning, and applying knowledge across various tasks with human-like proficiency. Unlike today’s narrow AI systems that excel at tasks such as image recognition or language translation, AGI can tackle any intellectual challenge a human can, demonstrating a deep comprehension of diverse disciplines. This technology could transform medical practice, empowering machines to reliably synthesize large amounts of clinical data on a patient’s condition and interpret complex medical problems.
New FDA Policies Could Limit the Full Value of AI in Medicine.
Gottlieb S.
JAMA Health Forum. 2025 Feb 7;6(2):e250289. doi: 10.1001/jamahealthforum.2025.0289. PMID: 39913129. - One challenge will be how the US Food and Drug Administration (FDA) views these tools—its recent changes to policies related to the regulation of AI have added new uncertainties. Artificial intelligence tools with advanced analytical capabilities used in clinical practice, especially tools that synthesize complex clinical information from distinct sources, may automatically be classified as medical devices, regardless of their intended use. This may be particularly true when AI is integrated into electronic medical record (EMR) software, allowing AI to generate insights that might otherwise go unnoticed by clinicians. Such classification, however, could be at odds with the original intent of laws that were designed to regulate digital health tools based on their clinical use rather than on their analytical sophistication alone, or the sources of clinical data that they rely on.
New FDA Policies Could Limit the Full Value of AI in Medicine.
Gottlieb S.
JAMA Health Forum. 2025 Feb 7;6(2):e250289. doi: 10.1001/jamahealthforum.2025.0289. PMID: 39913129. - However, a high-value application of AI and AGI in medicine hinges on their seamless integration into EMRs, accessing and synthesizing diverse data. It may be difficult to develop these CDSS tools separately from the EMR and purchase them as distinct modules without limiting their inherent utility. If these tools are classified as medical devices merely because they draw from multiple data sources or possess analytical capabilities that are so comprehensive and intelligent that clinicians are likely to accept their analyses in full, then nearly any AI tool embedded in an EMR could fall under regulation. The risk is that EMR developers may attempt to circumvent regulatory uncertainty by omitting these features from their software. This could deny health care clinicians access to AI tools that have the potential to transform the productivity and safety of medical care.
New FDA Policies Could Limit the Full Value of AI in Medicine.
Gottlieb S.
JAMA Health Forum. 2025 Feb 7;6(2):e250289. doi: 10.1001/jamahealthforum.2025.0289. PMID: 39913129. - A solution lies in returning to the intent of the 21st Century Cures Act and the policies advanced from 2017 to 2019. The intent was to regulate CDSS based on how the data analysis is presented to health care clinicians instead of focusing on how clinicians would use the information to inform their judgment. If these AI tools are designed to augment the information available to clinicians and do not provide autonomous diagnoses or treatment decisions, they should not be subjected to premarket review. The FDA could allow EMR providers to come to market with these tools as long as they meet FDA criteria for how they are designed and validated. Then, by drawing on real-world evidence of these systems in action in the postmarket setting, the agency can verify that they genuinely enhance the quality of medical decision-making.
New FDA Policies Could Limit the Full Value of AI in Medicine.
Gottlieb S.
JAMA Health Forum. 2025 Feb 7;6(2):e250289. doi: 10.1001/jamahealthforum.2025.0289. PMID: 39913129. - “Artificial intelligence has an inherent ability to synthesize complex information streams and deliver enhanced analyses or recommendations that might otherwise evade notice. That aptitude alone should not classify them as devices.”
New FDA Policies Could Limit the Full Value of AI in Medicine.
Gottlieb S.
JAMA Health Forum. 2025 Feb 7;6(2):e250289. doi: 10.1001/jamahealthforum.2025.0289. PMID: 39913129.
- Our research concludes that the proposed CAD (computer-aided diagnosis) system for pancreatic cancer detection and classification using deep learning is effective, achieving good performance results. The proposed system aims to address the limitations of the manual identification of pancreatic tumors by radiologists, which is challenging and time-consuming due to the complex nature of CT scan images. The objective of the work is to apply a deep learning model to create a four-stage framework for the preprocessing, segmentation, detection, and classification of pancreatic cancers. The potential for this discovery to transform early pancreatic cancer detection and classification makes it significant. The suggested CAD system can greatly increase diagnostic efficiency and accuracy by automating the tumor identification and categorization process. This might result in early detection and the potential to save many lives worldwide.
Automated CAD system for early detection and classification of pancreatic cancer using deep learning model
Nadeem A et al.
PLOS ONE | https://doi.org/10.1371/journal.pone.0307900 January 3, 2025 
Integrating OMICS-based platforms and analytical tools for diagnosis and management of pancreatic cancer: a review.
Sousa P, Silva L, Câmara JS, Guedes de Pinho P, Perestrelo R.
Mol Omics. 2024 Dec 23. doi: 10.1039/d4mo00187g.
Integrating OMICS-based platforms and analytical tools for diagnosis and management of pancreatic cancer: a review.
Sousa P, Silva L, Câmara JS, Guedes de Pinho P, Perestrelo R.
Mol Omics. 2024 Dec 23. doi: 10.1039/d4mo00187g.
Integrating OMICS-based platforms and analytical tools for diagnosis and management of pancreatic cancer: a review.
Sousa P, Silva L, Câmara JS, Guedes de Pinho P, Perestrelo R.
Mol Omics. 2024 Dec 23. doi: 10.1039/d4mo00187g.
Integrating OMICS-based platforms and analytical tools for diagnosis and management of pancreatic cancer: a review.
Sousa P, Silva L, Câmara JS, Guedes de Pinho P, Perestrelo R.
Mol Omics. 2024 Dec 23. doi: 10.1039/d4mo00187g.- To achieve the best possible outcomes for PaC patients, the development of new tests that could improve early diagnosis of the disease is essential. In this context, multi-OMICs approaches have emerged as a promising tool capable of revolutionizing both the early diagnosis and treatment of PaC. These integrative approaches not only have the potential to improve patient survival rates and quality of life, but also offer a personalized perspective for disease management. Researchers can develop more accurate predictive models for pre-diagnosis, and in depth analysis of molecular profiles allows the identification of patient subgroups with specific characteristics, paving the way for targeted and more effective and specific therapies.
Integrating OMICS-based platforms and analytical tools for diagnosis and management of pancreatic cancer: a review.
Sousa P, Silva L, Câmara JS, Guedes de Pinho P, Perestrelo R.
Mol Omics. 2024 Dec 23. doi: 10.1039/d4mo00187g. - “Similar improvements were observed when CA19-9 was combined with additional serum biomarkers, such as micro- RNAs. In a study of blood samples from 35 PaC patients and 15 CTLs, expression levels of selected microRNAs (miRNAs) and serum CA19-9 concentrations were determined by quantitative real-time reverse transcription-polymerase chain reaction (qRTPCR) and electrochemiluminescence immunoassay, respectively. Compared to CTLs, the levels of three miRNAs (miR-22- 3p, miR-642b-3p and miR-885-5p) were significantly higher in PaC patients, even in those with early-stage disease (IB and IIB). A panel of six miRNAs (let-7b-5p, miR-192-5p, miR-19a-3p, miR- 19b-3p, miR-223-3p, and miR-25-3p) together with serum miR- 25 in combination with CA19-9 and miR-17-5p methylation, showed superior diagnostic performance compared to CA19-9 or CEA.”
Integrating OMICS-based platforms and analytical tools for diagnosis and management of pancreatic cancer: a review.
Sousa P, Silva L, Câmara JS, Guedes de Pinho P, Perestrelo R.
Mol Omics. 2024 Dec 23. doi: 10.1039/d4mo00187g. - “There is also growing interest in the use of new approaches to detect potential biomarkers, highlighting the liquid biopsy, one of the most research hotspots that isolates the cancer-derived components from patients, including CTC, ctDNA, miRNA, lncRNA,99 which are present in body fluids such as blood, urine, and saliva. In addition, the development of modern technologies, including artificial intelligence (AI), can play a major role in the initial qualitative interpretation of cancer imaging, the prediction of clinical outcomes, and also the assessment of the impact of disease and treatment on adjacent organs. Particularly, the full integration of AI, molecular biomarkers, and complex intermolecular networks is expected to be a turning point in digital healthcare, ultimately improving personalized diagnostics and patient care.”
Integrating OMICS-based platforms and analytical tools for diagnosis and management of pancreatic cancer: a review.
Sousa P, Silva L, Câmara JS, Guedes de Pinho P, Perestrelo R.
Mol Omics. 2024 Dec 23. doi: 10.1039/d4mo00187g. - The synergy between AI, advanced biosensors and bioinformatics holds great promise for overcoming current limitations in cancer diagnosis. AI can identify subtle patterns and early biomarkers by analyzing large volumes of multi-OMICs data and advanced imaging, and continuously learning from clinical data to predict disease progression.101 Nanotechnology-based biosensors detect biomarkers at extremely low concentrations, such as ctDNA and miRNA, with high sensitivity and specificity, enabling rapid, accessible diagnostics and reducing reliance on invasive methods.102 By integrating and interpreting this complex data, bioinformatics can identify diagnostic clues and enable tailored treatments that improve efficacy and minimize side effects. This technological synergy has the potential to revolutionize oncology by enabling earlier detection, more precise treatment and improved quality of life for patients.
Integrating OMICS-based platforms and analytical tools for diagnosis and management of pancreatic cancer: a review.
Sousa P, Silva L, Câmara JS, Guedes de Pinho P, Perestrelo R.
Mol Omics. 2024 Dec 23. doi: 10.1039/d4mo00187g. - In parallel with these advances, analytical tools have influenced the study and management of PaC. MS, including high-resolution LC–MS/MS, MALDI Imaging MS, and GC–MS, allows in-depth profiling of proteins, metabolites, and their spatial changes in tumour tissue. Next-generation sequencing, including whole genome/exome sequencing and single-cell RNA sequencing, reveals the genetic landscape and tumour heterogeneity. Advanced imaging techniques, such as multiplex immunohistochemistry and fluorescent in situ hybridization, provide insight into cellular interactions and specific genetic alterations. Integrating this data with sophisticated bioinformatics approaches enables the identification of more precise biomarkers for early detection and the development of personalised therapies. As a result, these tools are significantly improving our understanding of PaC biology, potentially leading to more accurate diagnosis and improved patient outcomes.
Integrating OMICS-based platforms and analytical tools for diagnosis and management of pancreatic cancer: a review.
Sousa P, Silva L, Câmara JS, Guedes de Pinho P, Perestrelo R.
Mol Omics. 2024 Dec 23. doi: 10.1039/d4mo00187g. - Frequently, the pancreas appears morphologically normal during the pre-diagnostic phase, resulting in missed diagnoses when the disease is still in its early, treatable stages. This highlights the critical necessity of imaging-based biomarkers to improve early detection of sporadic PDAC in high-risk groups. While the pancreas may seem normal to the human eye in these early stages, AI models, such as ML algorithms analyzing radiomic features (e.g., shape, intensity, texture) and DL models like convolutional neural networks (CNNs), excel at processing complex imaging data. These tools can identify subtle patterns that often escape traditional visual assessments.”
AI-Driven insights in pancreatic cancer imaging: from pre-diagnostic detection to prognostication.
Antony A, Mukherjee S, Bi Y, Collisson EA, et al.
Abdom Radiol (NY). 2024 Dec 30. doi: 10.1007/s00261-024-04775-x. Epub ahead of print. PMID: 39738571. - “AI-powered tools for pancreas segmentation enable extraction of critical biomarkers for early PDAC detection at the asymptomatic stage on pre-diagnostic CTs. The latter are defined as incidental CTs conducted for unrelated clinical indications between 3 and 36 months before a clinical PDAC diagnosis. They are typically interpreted as negative for PDAC during routine clinical evaluations and confirmed as such during data curation . It is crucial to distinguish pre-diagnostic CTs from diagnostic CTs where a mass is present but missed during routine interpretation. The latter are referred to as diagnostic CTs with missed cancer. ”
AI-Driven insights in pancreatic cancer imaging: from pre-diagnostic detection to prognostication.
Antony A, Mukherjee S, Bi Y, Collisson EA, et al.
Abdom Radiol (NY). 2024 Dec 30. doi: 10.1007/s00261-024-04775-x. Epub ahead of print. PMID: 39738571. - “The significant differences in radiomic signatures between scans with healthy pancreas and pre-diagnostic cases, acquired 6 months to 3 years before PDAC diagnosis with no visible tumors, were highlighted in a recent study . This analysis, conducted on an internal dataset of 66 patients (22 healthy controls, 22 pre-diagnostic, and 22 diagnostic CTs), identified specific radiomic features that clearly distinguished healthy from pre-diagnostic groups. A method using a step-by-step feature selection process and a simple classification model was developed, achieving a mean accuracy of 86% when categorizing scans as either healthy or pre-diagnostic, validated on an external dataset of 28 scans (14 in each group). Furthermore, the study also demonstrated incremental trends in these features as pre-diagnostic cases progressed toward diagnostic scans.”
AI-Driven insights in pancreatic cancer imaging: from pre-diagnostic detection to prognostication.
Antony A, Mukherjee S, Bi Y, Collisson EA, et al.
Abdom Radiol (NY). 2024 Dec 30. doi: 10.1007/s00261-024-04775-x. Epub ahead of print. PMID: 39738571. - “In addition to detection of subtle, pre-diagnostic imaging signatures within the pancreas, AI can also monitor systemic changes linked to early PDAC, a disease with well-known sequalae in many organ systems. These changes include variations in body composition and metabolic function that contribute to conditions like cancer-induced cachexia, which leads to muscle loss, with or without fat depletion . Cachexia can begin even in the preclinical stages of the disease, when precursor lesions are present . A recent study evaluated longitudinal changes in body composition, specifically, subcutaneous adipose tissue (SAT), visceral adipose tissue (VAT), skeletal muscle area (SMA), and vertebral bone area and density, to assess the potential of extra-pancreatic imaging markers for pre-diagnostic PDAC detection. ”
AI-Driven insights in pancreatic cancer imaging: from pre-diagnostic detection to prognostication.
Antony A, Mukherjee S, Bi Y, Collisson EA, et al.
Abdom Radiol (NY). 2024 Dec 30. doi: 10.1007/s00261-024-04775-x. Epub ahead of print. PMID: 39738571. - “Recent innovations aim to address the limitations of DL-based segmentation. A notable study developed a semi-automated, bounding-box CNN for PDAC segmentation, designed to focus on peri-tumoral regions rather than the entire pancreas. By limiting the input to this focused area, the CNN was able to significantly improve segmentation performance. This model was trained on the largest dataset reported to date—1,151 portal venous-phase CTs from treatment-naïve patients with biopsy-confirmed PDAC. The bounding-box CNN achieved a high DSC of 0.84 on the internal test subset and demonstrated excellent generalizability across two public datasets: Medical Segmentation Decathlon (MSD) (DSC: 0.82) and The Cancer Imaging Archive (TCIA) (DSC: 0.84) . The model’s ability to generalize across public datasets further validates its potential for clinical adoption. ”
AI-Driven insights in pancreatic cancer imaging: from pre-diagnostic detection to prognostication.
Antony A, Mukherjee S, Bi Y, Collisson EA, et al.
Abdom Radiol (NY). 2024 Dec 30. doi: 10.1007/s00261-024-04775-x. Epub ahead of print. PMID: 39738571. - ”Integrating multi-domain data from various sources, such as imaging and clinico-pathologic variables, has emerged as a powerful method for predicting patient outcomes. By merging diverse data sources, AI models can extract complex, multidimensional patterns, offering a more enriched dataset than unimodal approaches. For instance, recent studies that integrated radiomic features with clinical variables such as age, CA19-9 levels, tumor morphology, and metastatic status demonstrated superior performance in survival prediction compared to models relying on radiomics or clinical features alone .” AI-Driven insights in pancreatic cancer imaging: from pre-diagnostic detection to prognostication.
Antony A, Mukherjee S, Bi Y, Collisson EA, et al.
Abdom Radiol (NY). 2024 Dec 30. doi: 10.1007/s00261-024-04775-x. Epub ahead of print. PMID: 39738571. - “Radiomic signatures have also been evaluated for their utility in predicting early recurrence in post-surgical patients. A recent model achieved an AUC of 0.73 in the training dataset and 0.67 in the validation dataset for predicting early recurrence. Multivariate analysis confirmed that radiomic signatures were independent predictors of early recurrence, supporting the robustness of AI-driven approaches in this context . Similarly, radiomics-based models have demonstrated high performance in predicting liver and lymph node metastasis, with AUCs of 0.7 and 0.85, respectively.”
AI-Driven insights in pancreatic cancer imaging: from pre-diagnostic detection to prognostication.
Antony A, Mukherjee S, Bi Y, Collisson EA, et al.
Abdom Radiol (NY). 2024 Dec 30. doi: 10.1007/s00261-024-04775-x. Epub ahead of print. PMID: 39738571. - “Publicly available datasets, vital for external validation, often lack quality and completeness, further limiting AI implementation in clinical practice. For instance, a recent study [66] assessing these datasets identified substantial quality gaps, including suboptimal imaging, incomplete annotations, and inherent biases. A significant proportion of CTs lacked essential details about the tumor histopathology, and approximately 25% include biliary stents. The presence of biliary stents introduces bias in AI models, as these models may erroneously associate the presence of stents with a PDAC diagnosis.”
AI-Driven insights in pancreatic cancer imaging: from pre-diagnostic detection to prognostication.
Antony A, Mukherjee S, Bi Y, Collisson EA, et al.
Abdom Radiol (NY). 2024 Dec 30. doi: 10.1007/s00261-024-04775-x. Epub ahead of print. PMID: 39738571. - “AI has shown great potential in transforming the early detection, diagnosis, and prognostication of PDAC, often outperforming radiologists by identifying subtle imaging patterns imperceptible to the human eye. This capability can improve patient outcomes by detecting PDAC at treatable stages. Beyond early detection, AI aids in prognostication by predicting patient outcomes, early recurrence, and metastasis risk with precision. Integrating non-pixel-based data can further enhance predictive accuracy, advancing precision medicine. However, challenges like limited pre-diagnostic data and model generalizability must be addressed for broader clinical adoption, requiring collaborative efforts to standardize data and expand high-quality datasets. As AI evolves, it is poised to enhance diagnostic accuracy, support personalized treatment planning, and improve survival rates, revolutionizing PDAC detection and treatment.”
AI-Driven insights in pancreatic cancer imaging: from pre-diagnostic detection to prognostication.
Antony A, Mukherjee S, Bi Y, Collisson EA, et al.
Abdom Radiol (NY). 2024 Dec 30. doi: 10.1007/s00261-024-04775-x. Epub ahead of print. PMID: 39738571. - “The WHO grading of pancreatic neuroendocrine neoplasms (PanNENs) is essential in patient management and an independent prognostic factor for patient survival. Radiomics features from CE-CT images hold promise for the outcome and tumor grade prediction. However, variations in reconstruction parameters can impact the predictive value of radiomics. 127 patients with histopathologically confirmed PanNENs underwent CT scans with filtered back projection (B20f) and iterative (I26f) reconstruction kernels. 3190 radiomic features were extracted from tumors and pancreatic volumes. Wilcoxon paired tests assessed the impact of reconstruction kernels and ComBat harmonization efficiency. SVM models were employed to predict tumor grade using the entire set of radiomics features or only those identified as harmonizable. The models’ performance was assessed on an independent dataset of 36 patients. Significant differences, after correction for multiple testing, were observed in 69% of features in the pancreatic volume and 51% in the tumor volume with B20f and I26f kernels. SVM models demonstrated accuracy ranging from 0.67 (95%CI: 0.50–0.81) to 0.83 (95%CI: 0.69–0.94) in distinguishing grade 1 cases from higher grades. Reconstruction kernels alter radiomics features and iterative kernel models trended towards higher performance. ComBat harmonization mitigates kernel impacts but addressing this effect is crucial in studies involving data from different kernels.”
Can CT Image Reconstruction Parameters Impact the Predictive Value of Radiomics Features in Grading Pancreatic Neuroendocrine Neoplasms?
Tixier F, Lopez-Ramirez F, Blanco A, Yasrab M, Javed AA, Chu LC, Fishman EK, Kawamoto S
Bioengineering (Basel). 2025 Jan 16;12(1):80. doi: 10.3390/bioengineering12010080. PMID: 39851354; PMCID: PMC11763079. - “ The models’ performance was assessed on an independent dataset of 36 patients. Significant differences, after correction for multiple testing, were observed in 69% of features in the pancreatic volume and 51% in the tumor volume with B20f and I26f kernels. SVM models demonstrated accuracy ranging from 0.67 (95%CI: 0.50–0.81) to 0.83 (95%CI: 0.69–0.94) in distinguishing grade 1 cases from higher grades. Reconstruction kernels alter radiomics features and iterative kernel models trended towards higher performance. ComBat harmonization mitigates kernel impacts but addressing this effect is crucial in studies involving data from different kernels.”
Can CT Image Reconstruction Parameters Impact the Predictive Value of Radiomics Features in Grading Pancreatic Neuroendocrine Neoplasms?
Tixier F, Lopez-Ramirez F, Blanco A, Yasrab M, Javed AA, Chu LC, Fishman EK, Kawamoto S
Bioengineering (Basel). 2025 Jan 16;12(1):80. doi: 10.3390/bioengineering12010080. PMID: 39851354; PMCID: PMC11763079. - “Radiomics analysis has emerged as a valuable tool in constructing prognostic and predictive models in oncology, leveraging the capability of radiomic features to capture underlying biological characteristics. Machine learning models based on radiomics features have demonstrated valuable clinical applications, supported by a growing body of evidence. Notably, these models have proven to be effective in applications such as predicting the histological grade of PanNENs in computed tomography (CT) images , offering guidance for follow-up and clinical decision-making. Preoperative tumor grading is essential for the effective clinical management of patients with PanNEN. However, biopsy-based techniques, while commonly used, are not ideal due to their invasive nature and the risk of misclassification due to tumor heterogeneity.”
Can CT Image Reconstruction Parameters Impact the Predictive Value of Radiomics Features in Grading Pancreatic Neuroendocrine Neoplasms?
Tixier F, Lopez-Ramirez F, Blanco A, Yasrab M, Javed AA, Chu LC, Fishman EK, Kawamoto S
Bioengineering (Basel). 2025 Jan 16;12(1):80. doi: 10.3390/bioengineering12010080. PMID: 39851354; PMCID: PMC11763079. - “However, the integration of radiomics into clinical practice necessitates a concerted effort to standardize reconstruction algorithms. This task is particularly challenging given the rapid advancements in scanner technologies, such as photon counting CT, which introduce new complexities for achieving harmonization in radiomics. Nevertheless, these technological shifts also present opportunities to enhance the utility of radiomics.”
Can CT Image Reconstruction Parameters Impact the Predictive Value of Radiomics Features in Grading Pancreatic Neuroendocrine Neoplasms?
Tixier F, Lopez-Ramirez F, Blanco A, Yasrab M, Javed AA, Chu LC, Fishman EK, Kawamoto S
Bioengineering (Basel). 2025 Jan 16;12(1):80. doi: 10.3390/bioengineering12010080. PMID: 39851354; PMCID: PMC11763079. - “Image reconstruction represents one of the numerous challenges for the clinical use of radiomics. To facilitate the translation into clinical practice, it is essential to provide a detailed description of all image processing steps, from data acquisition to modeling, and follow already established guidelines such as those from the IBSI In multicenter studies, various parameters, including CT manufacturer and acquisition settings, can vary and impact radiomic features. These additional sources of variability should be considered and must be carefully managed to harmonize images or radiomics features prior to modeling. Furthermore, when interpreting and generalizing radiomics findings across different centers, it is essential to understand precisely how data vary from the datasets used to develop the models.”
Can CT Image Reconstruction Parameters Impact the Predictive Value of Radiomics Features in Grading Pancreatic Neuroendocrine Neoplasms?
Tixier F, Lopez-Ramirez F, Blanco A, Yasrab M, Javed AA, Chu LC, Fishman EK, Kawamoto S
Bioengineering (Basel). 2025 Jan 16;12(1):80. doi: 10.3390/bioengineering12010080. PMID: 39851354; PMCID: PMC11763079. - “In this paper, we explored the influence of two soft tissue reconstruction kernels (I26f and B20f) on radiomics features and their predictive value for determining PanNET grades. Our findings indicate that a substantial number of features are biased by the reconstruction kernel, and I26f showed more promise than B20f for predicting PanNET grades. For studies employing mixed data arising from different reconstruction kernels, it is imperative to address this effect through harmonization techniques, such as ComBat, and by being cautious if using features not identified as harmonizable.”
Can CT Image Reconstruction Parameters Impact the Predictive Value of Radiomics Features in Grading Pancreatic Neuroendocrine Neoplasms?
Tixier F, Lopez-Ramirez F, Blanco A, Yasrab M, Javed AA, Chu LC, Fishman EK, Kawamoto S
Bioengineering (Basel). 2025 Jan 16;12(1):80. doi: 10.3390/bioengineering12010080. PMID: 39851354; PMCID: PMC11763079. - The tumor grade based on the WHO classification system is an independent prognostic factor for survival in patients with PanNENs. Also, the low-grade small PanNETs are indolent tumors with a good prognosis, and patients with small nonfunctioning PanNETs may undergo active surveillance or surgical resection. Therefore, pretreatment prediction of the PanNENs pathological tumor grade is important in determining prognosis and helps to guide the management of patients.
Can CT Image Reconstruction Parameters Impact the Predictive Value of Radiomics Features in Grading Pancreatic Neuroendocrine Neoplasms?
Tixier F, Lopez-Ramirez F, Blanco A, Yasrab M, Javed AA, Chu LC, Fishman EK, Kawamoto S.
Bioengineering (Basel). 2025 Jan 16;12(1):80. - “The 2017 WHO classification system describes two categories of PanNENs: well-differentiated pancreatic neuroendocrine tumors (PanNETs) and poorly differentiated pancreatic neuroendocrine carcinoma (PanNECs). PanNETs are well-differentiated tumors with minimal to moderate atypia and lack of necrosis and express intense synaptophysin or chromogranin. A positivity is classified as grade 1, 2, or 3 based on the mitotic index and the Ki-67 index PanNECs are tumors with high mitotic index and Ki-67 index and are characterized by poorly differentiated tumors consisting of atypical cells with substantial necrosis that are faintly positive for neuroendocrine markers. The tumor grade based on the WHO classification system is an independent prognostic factor for survival in patients with PanNENs. Also, the low-grade small PanNETs are indolent tumors with a good prognosis, and patients with small nonfunctioning PanNETs may undergo active surveillance or surgical resection. Therefore, pretreatment prediction of the PanNENs pathological tumor grade is important in determining prognosis and helps to guide the management of patient.”
Can CT Image Reconstruction Parameters Impact the Predictive Value of Radiomics Features in Grading Pancreatic Neuroendocrine Neoplasms?
Tixier F, Lopez-Ramirez F, Blanco A, Yasrab M, Javed AA, Chu LC, Fishman EK, Kawamoto S.
Bioengineering (Basel). 2025 Jan 16;12(1):80. - “In this study, we assessed the impact of soft tissue image reconstruction kernels on the radiomics features, explored the possibility of correcting for this effect using ComBat harmonization, and evaluated the predictive value of the radiomic features from images reconstructed with B20f and I26f to distinguish between WHO grade 1 and higher grade PanNENs, including grade 2 or 3 PanNETs and PanNECs. The primary objective was to investigate the reconstruction variability to provide valuable insights to improve the generalizability of PanNENs grading models based on radiomics. However, the results on feature robustness to reconstruction kernel and ComBat harmonization should extend to other radiomic models based on contrast CT features obtained from images reconstructed with iterative or filtered back projection soft tissue reconstruction kernels.”
Can CT Image Reconstruction Parameters Impact the Predictive Value of Radiomics Features in Grading Pancreatic Neuroendocrine Neoplasms?
Tixier F, Lopez-Ramirez F, Blanco A, Yasrab M, Javed AA, Chu LC, Fishman EK, Kawamoto S.
Bioengineering (Basel). 2025 Jan 16;12(1):80. - “Radiomics converts imaging data into high-dimensional features that can be used to characterize spatial heterogeneity inherent in disease processes. The features of radiomics can be classified into signal intensity, shape, and texture. Signal intensity (first-order) features are derived from histograms of individual voxel signal intensities, providing measures of central tendency and shape of the distribution. Shape features are extracted from the three-dimensional surface of the region of interest. Texture features are calculated in three dimensions, considering the correlation of signal intensities of adjacent voxels. In addition, feature extraction may be performed after applying a secondary filter, such as a wavelet or Gaussian filter.”
Pancreatic ductal adenocarcinoma staging: a narrative review of radiologic techniques and advances.
Chu LC, Fishman EK
Int J Surg. 2024 Oct 1;110(10):6052-6063. - “AI can theoretically function as ‘second readers’ to improve radiologists’ sensitivity in the detection of small tumors, which potentially can be cured with surgical resection. A preliminary study by Liu et al. showed promising results suggesting that DL could accurately differentiate CT scans of patients with PDAC from CT scans of healthy controls. More recently, Chen et al. developed a DL tool that differentiated CT scans of patients with PDAC vs. healthy controls with 89.9% sensitivity, 95.9% specificity, and 93.4% accuracy in the local test set. They validated this DL tool on a Taiwanese nationwide external validation set and achieved 89.7% sensitivity, 92.8% specificity, and 91.4% accuracy. Also, Park et al. developed a different DL tool that achieved high sensitivity comparable to radiologists in the detection of not only pancreatic solid masses (98–100%) but also cystic masses 1.0 cm or larger (sensitivity 92–93%), bringing us closer to a universal pancreatic neoplasm detector.”
Pancreatic ductal adenocarcinoma staging: a narrative review of radiologic techniques and advances.
Chu LC, Fishman EK
Int J Surg. 2024 Oct 1;110(10):6052-6063. - “Preliminary studies using emerging technologies such as advanced visualization and AI have revealed the potential of these tools to improve the initial diagnosis and staging of patients with PDAC. However, there remain several limitations. Most of these studies have been single-center retrospective studies, and their promising results should be validated in future multicenter prospective studies. Secondly, one of the major criticisms of AI is its ‘blackbox’ nature, making it difficult for clinicians to decipher the rationale behind AI predictions. Explainable or ‘glassbox’ AI is an active area of research that aims to render AI models more easily understandable and may help improve their clinical acceptance. Thirdly, these tools should be integrated seamlessly into the workflow to ensure widespread clinical implementation.”
Pancreatic ductal adenocarcinoma staging: a narrative review of radiologic techniques and advances.
Chu LC, Fishman EK
Int J Surg. 2024 Oct 1;110(10):6052-6063. - “Radiology plays a significant role in the initial diagnosis and staging of patients with PDAC, triaging patients with resectable disease, and determining treatment response to neoadjuvant chemotherapy and radiation. CT is the most used radiologic modality for PDAC staging, with MRI and PET/CT usually reserved as problem-solving tools. Current challenges in staging include preoperative diagnosis of lymph node metastases, subtle liver and peritoneal metastases, and R0 resection following neoadjuvant therapy. Artificial intelligence offers the potential of earlier disease diagnosis at the localized disease stage and prognostic radiologic biomarkers to optimize patient management, which can help improve patient outcomes.”
Pancreatic ductal adenocarcinoma staging: a narrative review of radiologic techniques and advances.
Chu LC, Fishman EK
Int J Surg. 2024 Oct 1;110(10):6052-6063.
- ”This review provides a comprehensive overview of the current landscape in pancreatic cancer (PC) screening, diagnosis, and early detection. This emphasizes the need for targeted screening in high-risk groups, particularly those with familial predispositions and genetic mutations, such as BRCA1, BRCA2, and PALB2. This review highlights the sporadic nature of most PC cases and significant risk factors, including smoking, alcohol consumption, obesity, and diabetes. Advanced imaging techniques, such as Endoscopic Ultrasound (EUS) and Contrast-Enhanced Harmonic Imaging (CEH-EUS), have been discussed for their superior sensitivity in early detection. This review also explores the potential of novel biomarkers, including those found in body fluids, such as serum, plasma, urine, and bile, as well as the emerging role of liquid biopsy technologies in analyzing circulating tumor DNA (ctDNA), circulating tumor cells (CTCs), and exosomes. AI-driven approaches, such as those employed in Project Felix and CancerSEEK, have been highlighted for their potential to enhance early detection through deep learning and biomarker discovery.”
Informatics strategies for early detection and risk mitigation in pancreatic cancer patients.
Jin D, Khan NU, Gu W, Lei H, Goel A, Chen T.
Neoplasia. 2025 Jan 21;60:101129. doi: 10.1016/j.neo.2025.101129. - ”Pancreatic ductal adenocarcinoma (PDAC), which constitutes approximately 90% of pancreatic cancer cases, is particularly notorious for its late-stage presentation. The lack of early symptoms and the deep anatomical location of the pancreas contribute to the challenge of timely diagnosis. Risk factors include smoking, obesity, diabetes mellitus, and a family history of the disease, with smoking alone doubling this risk. Notably, new-onset diabetes in older individuals is both a symptom and risk factor for pancreatic cancer, highlighting the need for vigilant monitoring. The early detection of pancreatic cancer remains a significant hurdle due to the absence of standardized screening protocols and reliable biomarkers. ”
Informatics strategies for early detection and risk mitigation in pancreatic cancer patients.
Jin D, Khan NU, Gu W, Lei H, Goel A, Chen T.
Neoplasia. 2025 Jan 21;60:101129. doi: 10.1016/j.neo.2025.101129. - “Precursor lesions, including pancreatic intraepithelial neoplasia (PanIN), intraductal papillary mucinous neoplasms (IPMNs), and mucinous cystic neoplasms (MCNs), are often detected by imaging, but only a tiny fraction progress to high-grade neoplasia. Molecular evidence indicates that most PDACs originate from PanINs, which are not visible with current imaging techniques, thus complicating early detection. The extended timeline of progression from precursors to invasive PDAC further challenges timely intervention and effective management. ”
Informatics strategies for early detection and risk mitigation in pancreatic cancer patients.
Jin D, Khan NU, Gu W, Lei H, Goel A, Chen T.
Neoplasia. 2025 Jan 21;60:101129. doi: 10.1016/j.neo.2025.101129. - “The lifetime risk of pancreatic ductal adenocarcinoma (PDAC) is approximately 1.7% in an average individual, but several clinical factors can significantly increase this risk. Chronic pancreatitis, mainly hereditary or recurrent, contributes to PDAC risk through ongoing tissue injury, inflammation, and DNA damage. However, recent estimates suggest a risk reduction to below 10% from earlier estimates as high as 70%. Acute pancreatitis, on the other hand, is linked to a heightened risk of PDAC within one year of the episode, likely due to tumor-induced obstruction. Smoking doubles the risk of PDAC, with meta-analyses indicating that smoking accounts for 11%–32% of pancreatic cancer cases. Heavy alcohol consumption also modestly increases PDAC risk, whereas light drinking does not. Obesity is associated with an elevated risk of PDAC, with studies showing a relative risk of 1.72 for individuals with a body mass index (BMI) of over 30. ”
Informatics strategies for early detection and risk mitigation in pancreatic cancer patients.
Jin D, Khan NU, Gu W, Lei H, Goel A, Chen T.
Neoplasia. 2025 Jan 21;60:101129. doi: 10.1016/j.neo.2025.101129. - “Intraductal papillary mucinous neoplasms (IPMNs) and mucinous cystic neoplasms (MCNs) of the pancreas offer opportunities for early detection of premalignant lesions, potentially targeting early detection strategies. However, their general applicability to pancreatic ductal adenocarcinoma (PDAC) is limited because premalignant cystic lesions serve as precursors for PDAC in only approximately 15% of cases. Despite this, IPMNs, particularly those in the main pancreatic duct, may harbor PDACs, thus warranting specialized surveillance or surgical resection based on lesion size and morphology. Additionally, individuals undergoing surveillance for IPMNs who later develop PDACs separate from IPMNs are not uncommon, given that PanINs, often undetectable by imaging, are generally more abundant in resected pancreata than in IPMNs.”
Informatics strategies for early detection and risk mitigation in pancreatic cancer patients.
Jin D, Khan NU, Gu W, Lei H, Goel A, Chen T.
Neoplasia. 2025 Jan 21;60:101129. doi: 10.1016/j.neo.2025.101129. - “Some current examples of AI technology in this area include Project Felix at Johns Hopkins University. Researchers developed a deep learning system for detecting small pancreatic tumors using CT scan images. Project Felix has brought significant improvement in the detection of pancreatic cancer, especially the first and second stages, and this has been tested in a real-world setting through clinical trials among patients with pancreatic cancer. CancerSEEK , a microfluidic platform coated with nanoparticles for identifying genetic mutation and protein biomarkers, using AI for biomarker identification, can detect several types of cancer, including pancreatic cancer. The earlier researcher showed that through CancerSEEK PDACs could be diagnosed, especially at early stages, with high sensitivity and specificity.”
Informatics strategies for early detection and risk mitigation in pancreatic cancer patients.
Jin D, Khan NU, Gu W, Lei H, Goel A, Chen T.
Neoplasia. 2025 Jan 21;60:101129. doi: 10.1016/j.neo.2025.101129. - “Personalized medicine is another significant focus, where genomic and proteomic data are integrated to tailor diagnostic and therapeutic approaches for individual patients. This involves leveraging patient- specific data to predict disease risk, optimize treatment plans, and improve outcomes. Personalized diagnostic tools are being developed to match patients with targeted therapies based on their unique genetic profiles and tumor characteristics [134]. Furthermore, ongoing research on pancreatic cancer’s tumor microenvironment and immune landscape is expected to yield new diagnostic and therapeutic strategies. Understanding the interactions between pancreatic cancer cells and their surrounding stroma and the role of immune cells can lead to the development of novel biomarkers and targeted therapies that address the complexities of the disease [135]. The future of pancreatic cancer diagnosis lies in the convergence of advanced imaging technologies, molecular and genetic profiling, AI-driven analysis, and personalized approaches. These advancements are poised to improve early detection, enhance diagnostic accuracy, and ultimately lead to improved patient outcomes.”
Informatics strategies for early detection and risk mitigation in pancreatic cancer patients.
Jin D, Khan NU, Gu W, Lei H, Goel A, Chen T.
Neoplasia. 2025 Jan 21;60:101129. doi: 10.1016/j.neo.2025.101129. - “AI-driven tools such as Project Felix and CancerSEEK offer new possibilities for enhancing diagnostic accuracy, refining risk assessment, and personalizing treatment. This strategic application of informatics in pancreatic cancer represents a comprehensive approach to improve early detection and risk management. Continued research and innovation are crucial for overcoming the existing limitations and advancing the fight against this formidable disease, ultimately aiming to save lives and improve patient outcomes.”
Informatics strategies for early detection and risk mitigation in pancreatic cancer patients.
Jin D, Khan NU, Gu W, Lei H, Goel A, Chen T.
Neoplasia. 2025 Jan 21;60:101129. doi: 10.1016/j.neo.2025.101129. 
Informatics strategies for early detection and risk mitigation in pancreatic cancer patients.
Jin D, Khan NU, Gu W, Lei H, Goel A, Chen T.
Neoplasia. 2025 Jan 21;60:101129. doi: 10.1016/j.neo.2025.101129.
Informatics strategies for early detection and risk mitigation in pancreatic cancer patients.
Jin D, Khan NU, Gu W, Lei H, Goel A, Chen T.
Neoplasia. 2025 Jan 21;60:101129. doi: 10.1016/j.neo.2025.101129.
- • Radiology plays an important role in the initial diagnosis and staging of pancreatic cancer.
• CT is the preferred modality over MRI due to wider availability, greater consistency in image quality, and lower cost.
• Patients can be triaged into resectable, borderline resectable, and locally advanced based on tumor involvement of arteries and veins.
• Accuracy of diagnosis and staging critically depends on the imaging technique and experience of the radiologists.
• Artificial intelligence has the potential to function as ‘second readers’ to improve the detection of small early stage tumors and provide imaging biomarkers to predict patient prognosis. .
Pancreatic ductal adenocarcinoma staging: a narrative review of radiologic techniques and advances.
Chu LC, Fishman EK
Int J Surg. 2024 Oct 1;110(10):6052-6063. - PDACs classically present as hypoenhancing masses with associated pancreatic duct dilatation and glandular atrophy of the body and tail. Pancreatic head tumors can cause common bile duct dilatation in addition to pancreatic duct dilatation, also known as the ‘double duct sign’. Up to 20% of PDACs enhance to the same degree as the background pancreas, and this isoattenuating pattern is more commonly found with smaller ( ≤20 mm) tumors. These small isoattenuating tumors can be difficult to detect on CT; therefore, radiologists often rely on secondary signs of the pancreatic duct or common bile duct dilatation for tumor detection. MRI and PET/CT have reported sensitivities of 79.2 and 73.7% in the detection of isoattenuating tumors, respectively, and may aid in detecting suspected pancreatic tumors that are occult on CT.”
Pancreatic ductal adenocarcinoma staging: a narrative review of radiologic techniques and advances.
Chu LC, Fishman EK
Int J Surg. 2024 Oct 1;110(10):6052-6063. - “Both arterial and venous involvement are pivotal in determining resectability. Based on the National Comprehensive Cancer Network (NCCN) guidelines, tumors without arterial tumor contact or superior mesenteric vein (SMV) or portal vein (PV) tumor contact are considered resectable. Tumors with ≤ 180° contact with the SMV or PV without contour irregularity are also considered resectable. Arterial abutment of the celiac artery or superior mesenteric artery (SMA) (< 180°) is considered borderline resectable, whereas arterial encasement ( ≥ 180° ) is usually considered locally advanced.”
Pancreatic ductal adenocarcinoma staging: a narrative review of radiologic techniques and advances.
Chu LC, Fishman EK
Int J Surg. 2024 Oct 1;110(10):6052-6063. - “The reported accuracy in determining tumor resectability ranges from 73 to 87% for CT and 70 to 79% for MRI, although this may depend on radiologists’ experience. CT offers superior spatial resolution and is less susceptible to artifacts compared to MRI. Also, CT allows for greater confidence in the assessment of tumor–vascular relationships. MRI is a critical problem-solving tool in the characterization of indeterminate liver lesions, which influences staging localized vs. metastatic disease. PET lacks the spatial resolution critical for the staging of locoregional involvement and is not used routinely in staging.”
Pancreatic ductal adenocarcinoma staging: a narrative review of radiologic techniques and advances.
Chu LC, Fishman EK
Int J Surg. 2024 Oct 1;110(10):6052-6063. 
Pancreatic ductal adenocarcinoma staging: a narrative review of radiologic techniques and advances.
Chu LC, Fishman EK
Int J Surg. 2024 Oct 1;110(10):6052-6063.
Pancreatic ductal adenocarcinoma staging: a narrative review of radiologic techniques and advances.
Chu LC, Fishman EK
Int J Surg. 2024 Oct 1;110(10):6052-6063.- “Cinematic rendering can accentuate subtle texture changes and improve tumor conspicuity relative to traditional 2D images, 3D volume rendering, or maximum intensity projection images. Cinematic rendering may be able to enhance the visualization of spatial relationships among the tumor and adjacent vasculature, differentiating true tumor infiltration from simple proximity to vessels. This can potentially improve the assessment of resectability and assist in determining optimal vascular reconstruction options. Cinematic rendering vascular maps illustrate the major arteries and veins with exquisite detail and can highlight the presence of variant vascular anatomy that may increase the risk of complications, such as hemorrhage, ischemia, anastomotic leakage, or pseudoaneurysm formation.”
Pancreatic ductal adenocarcinoma staging: a narrative review of radiologic techniques and advances.
Chu LC, Fishman EK
Int J Surg. 2024 Oct 1;110(10):6052-6063. - “Radiomics converts imaging data into high-dimensional features that can be used to characterize spatial heterogeneity inherent in disease processes. The features of radiomics can be classified into signal intensity, shape, and texture. Signal intensity (first-order) features are derived from histograms of individual voxel signal intensities, providing measures of central tendency and shape of the distribution. Shape features are extracted from the three-dimensional surface of the region of interest. Texture features are calculated in three dimensions, considering the correlation of signal intensities of adjacent voxels. In addition, feature extraction may be performed after applying a secondary filter, such as a wavelet or Gaussian filter.”
Pancreatic ductal adenocarcinoma staging: a narrative review of radiologic techniques and advances.
Chu LC, Fishman EK
Int J Surg. 2024 Oct 1;110(10):6052-6063. - “AI can theoretically function as ‘second readers’ to improve radiologists’ sensitivity in the detection of small tumors, which potentially can be cured with surgical resection. A preliminary study by Liu et al. showed promising results suggesting that DL could accurately differentiate CT scans of patients with PDAC from CT scans of healthy controls. More recently, Chen et al. developed a DL tool that differentiated CT scans of patients with PDAC vs. healthy controls with 89.9% sensitivity, 95.9% specificity, and 93.4% accuracy in the local test set. They validated this DL tool on a Taiwanese nationwide external validation set and achieved 89.7% sensitivity, 92.8% specificity, and 91.4% accuracy. Also, Park et al. developed a different DL tool that achieved high sensitivity comparable to radiologists in the detection of not only pancreatic solid masses (98–100%) but also cystic masses 1.0 cm or larger (sensitivity 92–93%), bringing us closer to a universal pancreatic neoplasm detector.”
Pancreatic ductal adenocarcinoma staging: a narrative review of radiologic techniques and advances.
Chu LC, Fishman EK
Int J Surg. 2024 Oct 1;110(10):6052-6063. 
Early detection of pancreatic cancer in the era of precision medicine.
Ahmed TM, Kawamoto S, Lopez-Ramirez F, Yasrab M, Hruban RH, Fishman EK, Chu LC.
Abdom Radiol (NY). 2024 Oct;49(10):3559-3573. doi: 10.1007/s00261-024-04358-w. Epub 2024 May 18. PMID: 38761272.- “While the 5-year survival rate for patients with localized PDAC is a favorable 44%, the aggregated overall survival rate across all stages drops to 13% . This negative skew, and a major driver of the particularly poor outcomes of this disease, is because only 9% of patients present with localized disease at the time of diagnosis. Consequently, only a small fraction (10–20%) of patients may be eligible for surgical resection, which remains the only curative treatment option. Although earlier detection of disease is challenging given the nonspecific presentation and asymptomatic course of early PDAC, prior studies have demonstrated that the median survival of incidentally detected PDACs is 10 months greater than that of comparable symptomatic PDAC. Smaller tumors are also more amenable to surgical resection with resectability rates of up to 99% in tumors under 2 cm in size at the time of detection. Early detection therefore holds promise to significantly improve outcomes of pancreatic cancer.”
Early detection of pancreatic cancer in the era of precision medicine.
Ahmed TM, Kawamoto S, Lopez-Ramirez F, Yasrab M, Hruban RH, Fishman EK, Chu LC.
Abdom Radiol (NY). 2024 Oct;49(10):3559-3573. - “Amongst identified high-risk conditions, the International Cancer of the Pancreas Surveillance (CAPS) consortium consensus statement recommends screening for all patients with Peutz Jeghers syndrome (deleterious germline STK11 variants) and familial atypical multiple mole melanoma syndrome (deleterious germline CDKN2A variants) starting at 40 years of age. Carriers of deleterious variants in BRCA2, BRCA1, PALB2, ATM, MLH1, MSH2 or MSH6 with at least one affected first-degree blood relative are recommended to begin screening between 45 and 50 years of age It is estimated that these inherited genetic cancer syndromes account for 3–5% of cases of PDAC . Individuals with familial pancreatic cancer, defend as having two or more first degree relatives with pancreatic cancer, are also recommended to undergo screening. Patients with familial pancreatic cancer are estimated to constitute 5–10% of cases of PDAC.”
Early detection of pancreatic cancer in the era of precision medicine.
Ahmed TM, Kawamoto S, Lopez-Ramirez F, Yasrab M, Hruban RH, Fishman EK, Chu LC.
Abdom Radiol (NY). 2024 Oct;49(10):3559-3573. - ‘Liquid biopsy’, which refers to the sampling of biomarkers from bodily fluids, has garnered increased attention in recent years. Compared to traditional biopsies, liquid biopsies are minimally invasive, allow for convenient serial sampling and may better capture tumor heterogeneity—rendering them ideal screening tools. Currently, targets of liquid biopsy primarily include circulating tumor DNA (ctDNA), circulating tumor cells (CTCs), and exosomes. While CTCs and exosomes have been an exciting area of new research, ctDNA remains the most widely studied targets for liquid biopsy and hold greatest clinical utility.”
Early detection of pancreatic cancer in the era of precision medicine.
Ahmed TM, Kawamoto S, Lopez-Ramirez F, Yasrab M, Hruban RH, Fishman EK, Chu LC.
Abdom Radiol (NY). 2024 Oct;49(10):3559-3573. - “The KRAS gene has been demonstrated to be the most common somatically mutated gene in PDAC. Approximately 95% of PDAC harbors mutant KRAS alleles, and tests that can detect these mutations have demonstrated impressive sensitivity for early-stage detection, with a previous study demonstrating that up to a third of patients with low-stage PDAC had detectable KRAS or GNAS mutations in circulating DNA. A number of studies have also evaluated the combined performance of ctDNA with other biomarkers . Notably, Cohen et al. demonstrated that KRAS ctDNA detection combined with a protein biomarker assay consisting of CA19-9, CEA, hepatocyte growth factor (HGF) and osteopontin (OPN) increased the sensitivity compared to detecting KRAS alone in suggesting the diagnosis of PDAC .”
Early detection of pancreatic cancer in the era of precision medicine.
Ahmed TM, Kawamoto S, Lopez-Ramirez F, Yasrab M, Hruban RH, Fishman EK, Chu LC.
Abdom Radiol (NY). 2024 Oct;49(10):3559-3573. - “Emerging studies have also ventured beyond application of AI techniques as ‘second readers’ on diagnostic scans and investigated the ability of AI models to detect visually occult PDAC on prediagnostic images months before the cancer is discernable to the human eye. Korfatis et al. demonstrated that a deep learning model could classify PDAC and controls on CT images with AUC of 0.91 at a median lead time of 475 days before clinical diagnosis. Other similar studies on this have reported an accuracy of 89% to 100% at a lead time ranging from 3 to 36 months before clinical diagnosis. Augmentation of current screening protocols with such models that specialize in detecting preinvasive lesions, particularly in conjunction with blood-based biomarkers holds potential to further streamline noninvasive screening strategies.”
Early detection of pancreatic cancer in the era of precision medicine.
Ahmed TM, Kawamoto S, Lopez-Ramirez F, Yasrab M, Hruban RH, Fishman EK, Chu LC.
Abdom Radiol (NY). 2024 Oct;49(10):3559-3573. - “Perhaps the most significant challenge of screening for PDAC is that of balancing the sensitivity of screening programs with the considerable risks of pancreatic surgery. This is particularly pressing as up to nearly 70% of surgeries undergone by HRIs under surveillance for PDAC, were, in retrospect, unnecessary. IPMNs contribute to the complexity of this challenge with prior studies demonstrating that only 25% of branch duct IPMN and 66% of main-duct IPMN that are surgically resected harbor high-grade dysplasia or an associated invasive carcinoma. Enhanced preoperative evaluation of IPMN dysplasia is therefore crucial to avoid unnecessary surgeries and optimize current screening protocols. “
Early detection of pancreatic cancer in the era of precision medicine.
Ahmed TM, Kawamoto S, Lopez-Ramirez F, Yasrab M, Hruban RH, Fishman EK, Chu LC.
Abdom Radiol (NY). 2024 Oct;49(10):3559-3573. - “A significant barrier to early detection even in high-risk cohorts is that the pancreas often appears morphologically normal during the pre-diagnostic phase. Yet, the disease can progress rapidly from subclinical stages to widespread metastasis, undermining the effectiveness of screening. Recently, artificial intelligence (AI) applied to cross-sectional imaging has shown significant potential in identifying subtle, early-stage changes in pancreatic tissue that are often imperceptible to the human eye. Moreover, AI-driven imaging also aids in the discovery of prognostic and predictive biomarkers, essential for personalized treatment planning. ”
AI-Driven insights in pancreatic cancer imaging: from pre-diagnostic detection to prognostication.
Antony A, Mukherjee S, Bi Y, Collisson EA, et al.
Abdom Radiol (NY). 2024 Dec 30. doi: 10.1007/s00261-024-04775-x. Epub ahead of print. PMID: 39738571. - ” Beyond early detection, AI plays a crucial role in prognostication and treatment response assessment for PDAC. Furthermore, AI-driven quantitative imaging can continuously monitor treatment response, offering dynamic insights into how PDAC tumors evolve during therapy. By analyzing changes in tumor size, shape, and internal heterogeneity, AI tools provide dynamic assessments that can guide treatment adjustments. This capability is particularly vital in PDAC, where conventional imaging often lags in detecting therapeutic impact, leading to delays in optimizing treatment strategies. ”
AI-Driven insights in pancreatic cancer imaging: from pre-diagnostic detection to prognostication.
Antony A, Mukherjee S, Bi Y, Collisson EA, et al.
Abdom Radiol (NY). 2024 Dec 30. doi: 10.1007/s00261-024-04775-x. Epub ahead of print. PMID: 39738571. - “Early imaging findings of PDAC are often subtle and easily overlooked on CT scans. Many healthy individuals may also show these indirect signs, as they are not specific to PDAC. Evaluating such findings can be subjective, with low inter-reader agreement. Frequently, the pancreas appears morphologically normal during the pre-diagnostic phase, resulting in missed diagnoses when the disease is still in its early, treatable stages.”
AI-Driven insights in pancreatic cancer imaging: from pre-diagnostic detection to prognostication.
Antony A, Mukherjee S, Bi Y, Collisson EA, et al.
Abdom Radiol (NY). 2024 Dec 30. doi: 10.1007/s00261-024-04775-x. Epub ahead of print. PMID: 39738571. - Frequently, the pancreas appears morphologically normal during the pre-diagnostic phase, resulting in missed diagnoses when the disease is still in its early, treatable stages. This highlights the critical necessity of imaging-based biomarkers to improve early detection of sporadic PDAC in high-risk groups. While the pancreas may seem normal to the human eye in these early stages, AI models, such as ML algorithms analyzing radiomic features (e.g., shape, intensity, texture) and DL models like convolutional neural networks (CNNs), excel at processing complex imaging data. These tools can identify subtle patterns that often escape traditional visual assessments.”
AI-Driven insights in pancreatic cancer imaging: from pre-diagnostic detection to prognostication.
Antony A, Mukherjee S, Bi Y, Collisson EA, et al.
Abdom Radiol (NY). 2024 Dec 30. doi: 10.1007/s00261-024-04775-x. Epub ahead of print. PMID: 39738571. - “AI-powered tools for pancreas segmentation enable extraction of critical biomarkers for early PDAC detection at the asymptomatic stage on pre-diagnostic CTs. The latter are defined as incidental CTs conducted for unrelated clinical indications between 3 and 36 months before a clinical PDAC diagnosis. They are typically interpreted as negative for PDAC during routine clinical evaluations and confirmed as such during data curation . It is crucial to distinguish pre-diagnostic CTs from diagnostic CTs where a mass is present but missed during routine interpretation. The latter are referred to as diagnostic CTs with missed cancer. ”
AI-Driven insights in pancreatic cancer imaging: from pre-diagnostic detection to prognostication.
Antony A, Mukherjee S, Bi Y, Collisson EA, et al.
Abdom Radiol (NY). 2024 Dec 30. doi: 10.1007/s00261-024-04775-x. Epub ahead of print. PMID: 39738571. - “The significant differences in radiomic signatures between scans with healthy pancreas and pre-diagnostic cases, acquired 6 months to 3 years before PDAC diagnosis with no visible tumors, were highlighted in a recent study . This analysis, conducted on an internal dataset of 66 patients (22 healthy controls, 22 pre-diagnostic, and 22 diagnostic CTs), identified specific radiomic features that clearly distinguished healthy from pre-diagnostic groups. A method using a step-by-step feature selection process and a simple classification model was developed, achieving a mean accuracy of 86% when categorizing scans as either healthy or pre-diagnostic, validated on an external dataset of 28 scans (14 in each group). Furthermore, the study also demonstrated incremental trends in these features as pre-diagnostic cases progressed toward diagnostic scans.”
AI-Driven insights in pancreatic cancer imaging: from pre-diagnostic detection to prognostication.
Antony A, Mukherjee S, Bi Y, Collisson EA, et al.
Abdom Radiol (NY). 2024 Dec 30. doi: 10.1007/s00261-024-04775-x. Epub ahead of print. PMID: 39738571. - “In addition to detection of subtle, pre-diagnostic imaging signatures within the pancreas, AI can also monitor systemic changes linked to early PDAC, a disease with well-known sequalae in many organ systems. These changes include variations in body composition and metabolic function that contribute to conditions like cancer-induced cachexia, which leads to muscle loss, with or without fat depletion . Cachexia can begin even in the preclinical stages of the disease, when precursor lesions are present . A recent study evaluated longitudinal changes in body composition, specifically, subcutaneous adipose tissue (SAT), visceral adipose tissue (VAT), skeletal muscle area (SMA), and vertebral bone area and density, to assess the potential of extra-pancreatic imaging markers for pre-diagnostic PDAC detection. ”
AI-Driven insights in pancreatic cancer imaging: from pre-diagnostic detection to prognostication.
Antony A, Mukherjee S, Bi Y, Collisson EA, et al.
Abdom Radiol (NY). 2024 Dec 30. doi: 10.1007/s00261-024-04775-x. Epub ahead of print. PMID: 39738571. - “Another DL model, PANDA (Pancreatic Cancer Detection with Artificial Intelligence), utilizes pancreas segmentation masks that include lesions as the initial step in a three-stage algorithm. The second stage focuses on lesion detection, identifying both PDAC and non-PDAC lesions, and is followed by differential diagnosis if a lesion is detected. The model was trained on a dataset of 3,208 non-contrast CT scans, where pancreas and lesion annotations were mapped from paired contrast-enhanced CT scans through image registration. PANDA achieved an AUC of 0.987 (95% CI 0.975–0.996) for PDAC identification on an internal dataset of 291 patients (108 PDAC scans, 67 non- PDAC lesion scans, and 116 control scans). On an external validation dataset of 5,337 patients (2,737 PDAC scans, 932 non-PDAC lesion scans, and 1,668 control scans), the model demonstrated a sensitivity of 90.1% (95% CI 89.0–91.2%) for PDAC identification. It maintained sensitivity above 90% for detecting stage 1 and stage 2 tumors across both datasets. ”
AI-Driven insights in pancreatic cancer imaging: from pre-diagnostic detection to prognostication.
Antony A, Mukherjee S, Bi Y, Collisson EA, et al.
Abdom Radiol (NY). 2024 Dec 30. doi: 10.1007/s00261-024-04775-x. Epub ahead of print. PMID: 39738571. - “Recent innovations aim to address the limitations of DL-based segmentation. A notable study developed a semi-automated, bounding-box CNN for PDAC segmentation, designed to focus on peri-tumoral regions rather than the entire pancreas. By limiting the input to this focused area, the CNN was able to significantly improve segmentation performance. This model was trained on the largest dataset reported to date—1,151 portal venous-phase CTs from treatment-naïve patients with biopsy-confirmed PDAC. The bounding-box CNN achieved a high DSC of 0.84 on the internal test subset and demonstrated excellent generalizability across two public datasets: Medical Segmentation Decathlon (MSD) (DSC: 0.82) and The Cancer Imaging Archive (TCIA) (DSC: 0.84) . The model’s ability to generalize across public datasets further validates its potential for clinical adoption. ”
AI-Driven insights in pancreatic cancer imaging: from pre-diagnostic detection to prognostication.
Antony A, Mukherjee S, Bi Y, Collisson EA, et al.
Abdom Radiol (NY). 2024 Dec 30. doi: 10.1007/s00261-024-04775-x. Epub ahead of print. PMID: 39738571. - ”Integrating multi-domain data from various sources, such as imaging and clinico-pathologic variables, has emerged as a powerful method for predicting patient outcomes. By merging diverse data sources, AI models can extract complex, multidimensional patterns, offering a more enriched dataset than unimodal approaches. For instance, recent studies that integrated radiomic features with clinical variables such as age, CA19-9 levels, tumor morphology, and metastatic status demonstrated superior performance in survival prediction compared to models relying on radiomics or clinical features alone .”
AI-Driven insights in pancreatic cancer imaging: from pre-diagnostic detection to prognostication.
Antony A, Mukherjee S, Bi Y, Collisson EA, et al.
Abdom Radiol (NY). 2024 Dec 30. doi: 10.1007/s00261-024-04775-x. Epub ahead of print. PMID: 39738571. - “Radiomic signatures have also been evaluated for their utility in predicting early recurrence in post-surgical patients. A recent model achieved an AUC of 0.73 in the training dataset and 0.67 in the validation dataset for predicting early recurrence. Multivariate analysis confirmed that radiomic signatures were independent predictors of early recurrence, supporting the robustness of AI-driven approaches in this context . Similarly, radiomics-based models have demonstrated high performance in predicting liver and lymph node metastasis, with AUCs of 0.7 and 0.85, respectively.”
AI-Driven insights in pancreatic cancer imaging: from pre-diagnostic detection to prognostication.
Antony A, Mukherjee S, Bi Y, Collisson EA, et al.
Abdom Radiol (NY). 2024 Dec 30. doi: 10.1007/s00261-024-04775-x. Epub ahead of print. PMID: 39738571. - “Publicly available datasets, vital for external validation, often lack quality and completeness, further limiting AI implementation in clinical practice. For instance, a recent study [66] assessing these datasets identified substantial quality gaps, including suboptimal imaging, incomplete annotations, and inherent biases. A significant proportion of CTs lacked essential details about the tumor histopathology, and approximately 25% include biliary stents. The presence of biliary stents introduces bias in AI models, as these models may erroneously associate the presence of stents with a PDAC diagnosis.”
AI-Driven insights in pancreatic cancer imaging: from pre-diagnostic detection to prognostication.
Antony A, Mukherjee S, Bi Y, Collisson EA, et al.
Abdom Radiol (NY). 2024 Dec 30. doi: 10.1007/s00261-024-04775-x. Epub ahead of print. PMID: 39738571. - “AI has shown great potential in transforming the early detection, diagnosis, and prognostication of PDAC, often outperforming radiologists by identifying subtle imaging patterns imperceptible to the human eye. This capability can improve patient outcomes by detecting PDAC at treatable stages. Beyond early detection, AI aids in prognostication by predicting patient outcomes, early recurrence, and metastasis risk with precision. Integrating non-pixel-based data can further enhance predictive accuracy, advancing precision medicine. However, challenges like limited pre-diagnostic data and model generalizability must be addressed for broader clinical adoption, requiring collaborative efforts to standardize data and expand high-quality datasets. As AI evolves, it is poised to enhance diagnostic accuracy, support personalized treatment planning, and improve survival rates, revolutionizing PDAC detection and treatment.”
AI-Driven insights in pancreatic cancer imaging: from pre-diagnostic detection to prognostication.
Antony A, Mukherjee S, Bi Y, Collisson EA, et al.
Abdom Radiol (NY). 2024 Dec 30. doi: 10.1007/s00261-024-04775-x. Epub ahead of print. PMID: 39738571. 
Early detection of pancreatic cancer in the era of precision medicine.
Ahmed TM, Kawamoto S, Lopez-Ramirez F, Yasrab M, Hruban RH, Fishman EK, Chu LC.
Abdom Radiol (NY). 2024 Oct;49(10):3559-3573.- “While progress has been made in early detection strategies for PDAC, challenges persist. Screening protocols, liquid biopsy and AI techniques show promise in improving diagnoses, but their clinical adoption remains limited. Future integration of current divergent approaches into a fused multiomic framework remains necessary. Despite existing hurdles however, ongoing advances continue to make incremental progress, offering hope to enable early detection and to slowly but surely improve outcomes of this deadly disease.”
Early detection of pancreatic cancer in the era of precision medicine.
Ahmed TM, Kawamoto S, Lopez-Ramirez F, Yasrab M, Hruban RH, Fishman EK, Chu LC.
Abdom Radiol (NY). 2024 Oct;49(10):3559-3573. - “Another interesting avenue of AI developments that may be integrated into multifaceted frameworks are large language models (LLM). Medical documentation in electronic health records (EHR) is largely captured in unstructured text. LLMs such as GPT-4 (OpenAI) may therefore be uniquely suited to be applied to these datasets to streamline diagnostics and improve reporting. If securely integrated into EHRs, LLMs could swiftly mine EHRs to provide radiologists with clinically relevant information that may help frame imaging reports . An example of this could be to prompt LLMs to list all the risk factors a patient may have for pancreatic cancer and to stratify that patient’s risk based on an integration of the imaging findings with prior clinical features and demographics.”
Early detection of pancreatic cancer in the era of precision medicine.
Ahmed TM, Kawamoto S, Lopez-Ramirez F, Yasrab M, Hruban RH, Fishman EK, Chu LC.
Abdom Radiol (NY). 2024 Oct;49(10):3559-3573. - Recently, artificial intelligence (AI) applied to cross-sectional imaging has shown significant potential in identifying subtle, early-stage changes in pancreatic tissue that are often imperceptible to the human eye. Moreover, AI-driven imaging also aids in the discovery of prognostic and predictive biomarkers, essential for personalized treatment planning. This article uniquely integrates a critical discussion on AI’s role in detecting visually occult PDAC on pre-diagnostic imaging, addresses challenges of model generalizability, and emphasizes solutions like standardized datasets and clinical workflows. By focusing on both technical advancements and practical implementation, this article provides a forward-thinking conceptual framework that bridges current gaps in AI-driven PDAC research.
AI-Driven insights in pancreatic cancer imaging: from pre-diagnostic detection to prognostication.
Antony A, Mukherjee S, Bi Y, Collisson EA, et al.
Abdom Radiol (NY). 2024 Dec 30. doi: 10.1007/s00261-024-04775-x. Epub ahead of print. PMID: 39738571.
Kidney
- As artificial intelligence continues to shape the medical imaging landscape, its integration into the field of urologic oncology has led to impressive results. For prostate cancer diagnostics, machine learning has shown promise in refining clinically-significant lesion detection, with some success in deciphering ambiguous lesions on multiparametric MRI. For kidney cancer, radiomics has emerged as a valuable tool for better distinguishing between benign and malignant renal masses and predicting tumor behavior from CT or MRI scans. Meanwhile, in the arena of bladder cancer, there is a burgeoning emphasis on prediction of muscle invasive cancer and forecasting disease trajectory. However, many studies showing promise in these areas face challenges due to limited sample sizes and the need for broader external validation.
AI-powered radiomics: revolutionizing detection of urologic malignancies.
Gelikman, David G.a; Rais-Bahrami, et al.
Current Opinion in Urology 34(1):p 1-7, January 2024. - Radiomics integrated with artificial intelligence offers a pioneering approach to urologic oncology, ushering in an era of enhanced diagnostic precision and reduced invasiveness, guiding patient-tailored treatment plans. Researchers must embrace broader, multicentered endeavors to harness the full potential of this field.
AI-powered radiomics: revolutionizing detection of urologic malignancies.
Gelikman, David G.a; Rais-Bahrami, et al.
Current Opinion in Urology 34(1):p 1-7, January 2024.
Pancreas
- “The WHO grading of pancreatic neuroendocrine neoplasms (PanNENs) is essential in patient management and an independent prognostic factor for patient survival. Radiomics features from CE-CT images hold promise for the outcome and tumor grade prediction. However, variations in reconstruction parameters can impact the predictive value of radiomics. 127 patients with histopathologically confirmed PanNENs underwent CT scans with filtered back projection (B20f) and iterative (I26f) reconstruction kernels. 3190 radiomic features were extracted from tumors and pancreatic volumes. Wilcoxon paired tests assessed the impact of reconstruction kernels and ComBat harmonization efficiency. SVM models were employed to predict tumor grade using the entire set of radiomics features or only those identified as harmonizable. The models’ performance was assessed on an independent dataset of 36 patients. Significant differences, after correction for multiple testing, were observed in 69% of features in the pancreatic volume and 51% in the tumor volume with B20f and I26f kernels. SVM models demonstrated accuracy ranging from 0.67 (95%CI: 0.50–0.81) to 0.83 (95%CI: 0.69–0.94) in distinguishing grade 1 cases from higher grades. Reconstruction kernels alter radiomics features and iterative kernel models trended towards higher performance. ComBat harmonization mitigates kernel impacts but addressing this effect is crucial in studies involving data from different kernels.”
Can CT Image Reconstruction Parameters Impact the Predictive Value of Radiomics Features in Grading Pancreatic Neuroendocrine Neoplasms?
Tixier F, Lopez-Ramirez F, Blanco A, Yasrab M, Javed AA, Chu LC, Fishman EK, Kawamoto S
Bioengineering (Basel). 2025 Jan 16;12(1):80. doi: 10.3390/bioengineering12010080. PMID: 39851354; PMCID: PMC11763079. - “ The models’ performance was assessed on an independent dataset of 36 patients. Significant differences, after correction for multiple testing, were observed in 69% of features in the pancreatic volume and 51% in the tumor volume with B20f and I26f kernels. SVM models demonstrated accuracy ranging from 0.67 (95%CI: 0.50–0.81) to 0.83 (95%CI: 0.69–0.94) in distinguishing grade 1 cases from higher grades. Reconstruction kernels alter radiomics features and iterative kernel models trended towards higher performance. ComBat harmonization mitigates kernel impacts but addressing this effect is crucial in studies involving data from different kernels.”
Can CT Image Reconstruction Parameters Impact the Predictive Value of Radiomics Features in Grading Pancreatic Neuroendocrine Neoplasms?
Tixier F, Lopez-Ramirez F, Blanco A, Yasrab M, Javed AA, Chu LC, Fishman EK, Kawamoto S
Bioengineering (Basel). 2025 Jan 16;12(1):80. doi: 10.3390/bioengineering12010080. PMID: 39851354; PMCID: PMC11763079. - “Radiomics analysis has emerged as a valuable tool in constructing prognostic and predictive models in oncology, leveraging the capability of radiomic features to capture underlying biological characteristics. Machine learning models based on radiomics features have demonstrated valuable clinical applications, supported by a growing body of evidence. Notably, these models have proven to be effective in applications such as predicting the histological grade of PanNENs in computed tomography (CT) images , offering guidance for follow-up and clinical decision-making. Preoperative tumor grading is essential for the effective clinical management of patients with PanNEN. However, biopsy-based techniques, while commonly used, are not ideal due to their invasive nature and the risk of misclassification due to tumor heterogeneity.”
Can CT Image Reconstruction Parameters Impact the Predictive Value of Radiomics Features in Grading Pancreatic Neuroendocrine Neoplasms?
Tixier F, Lopez-Ramirez F, Blanco A, Yasrab M, Javed AA, Chu LC, Fishman EK, Kawamoto S
Bioengineering (Basel). 2025 Jan 16;12(1):80. doi: 10.3390/bioengineering12010080. PMID: 39851354; PMCID: PMC11763079. - “However, the integration of radiomics into clinical practice necessitates a concerted effort to standardize reconstruction algorithms. This task is particularly challenging given the rapid advancements in scanner technologies, such as photon counting CT, which introduce new complexities for achieving harmonization in radiomics. Nevertheless, these technological shifts also present opportunities to enhance the utility of radiomics.”
Can CT Image Reconstruction Parameters Impact the Predictive Value of Radiomics Features in Grading Pancreatic Neuroendocrine Neoplasms?
Tixier F, Lopez-Ramirez F, Blanco A, Yasrab M, Javed AA, Chu LC, Fishman EK, Kawamoto S
Bioengineering (Basel). 2025 Jan 16;12(1):80. doi: 10.3390/bioengineering12010080. PMID: 39851354; PMCID: PMC11763079. - “Image reconstruction represents one of the numerous challenges for the clinical use of radiomics. To facilitate the translation into clinical practice, it is essential to provide a detailed description of all image processing steps, from data acquisition to modeling, and follow already established guidelines such as those from the IBSI In multicenter studies, various parameters, including CT manufacturer and acquisition settings, can vary and impact radiomic features. These additional sources of variability should be considered and must be carefully managed to harmonize images or radiomics features prior to modeling. Furthermore, when interpreting and generalizing radiomics findings across different centers, it is essential to understand precisely how data vary from the datasets used to develop the models.”
Can CT Image Reconstruction Parameters Impact the Predictive Value of Radiomics Features in Grading Pancreatic Neuroendocrine Neoplasms?
Tixier F, Lopez-Ramirez F, Blanco A, Yasrab M, Javed AA, Chu LC, Fishman EK, Kawamoto S
Bioengineering (Basel). 2025 Jan 16;12(1):80. doi: 10.3390/bioengineering12010080. PMID: 39851354; PMCID: PMC11763079. - “In this paper, we explored the influence of two soft tissue reconstruction kernels (I26f and B20f) on radiomics features and their predictive value for determining PanNET grades. Our findings indicate that a substantial number of features are biased by the reconstruction kernel, and I26f showed more promise than B20f for predicting PanNET grades. For studies employing mixed data arising from different reconstruction kernels, it is imperative to address this effect through harmonization techniques, such as ComBat, and by being cautious if using features not identified as harmonizable.”
Can CT Image Reconstruction Parameters Impact the Predictive Value of Radiomics Features in Grading Pancreatic Neuroendocrine Neoplasms?
Tixier F, Lopez-Ramirez F, Blanco A, Yasrab M, Javed AA, Chu LC, Fishman EK, Kawamoto S
Bioengineering (Basel). 2025 Jan 16;12(1):80. doi: 10.3390/bioengineering12010080. PMID: 39851354; PMCID: PMC11763079. - The tumor grade based on the WHO classification system is an independent prognostic factor for survival in patients with PanNENs. Also, the low-grade small PanNETs are indolent tumors with a good prognosis, and patients with small nonfunctioning PanNETs may undergo active surveillance or surgical resection. Therefore, pretreatment prediction of the PanNENs pathological tumor grade is important in determining prognosis and helps to guide the management of patients.
Can CT Image Reconstruction Parameters Impact the Predictive Value of Radiomics Features in Grading Pancreatic Neuroendocrine Neoplasms?
Tixier F, Lopez-Ramirez F, Blanco A, Yasrab M, Javed AA, Chu LC, Fishman EK, Kawamoto S.
Bioengineering (Basel). 2025 Jan 16;12(1):80. - “The 2017 WHO classification system describes two categories of PanNENs: well-differentiated pancreatic neuroendocrine tumors (PanNETs) and poorly differentiated pancreatic neuroendocrine carcinoma (PanNECs). PanNETs are well-differentiated tumors with minimal to moderate atypia and lack of necrosis and express intense synaptophysin or chromogranin. A positivity is classified as grade 1, 2, or 3 based on the mitotic index and the Ki-67 index PanNECs are tumors with high mitotic index and Ki-67 index and are characterized by poorly differentiated tumors consisting of atypical cells with substantial necrosis that are faintly positive for neuroendocrine markers. The tumor grade based on the WHO classification system is an independent prognostic factor for survival in patients with PanNENs. Also, the low-grade small PanNETs are indolent tumors with a good prognosis, and patients with small nonfunctioning PanNETs may undergo active surveillance or surgical resection. Therefore, pretreatment prediction of the PanNENs pathological tumor grade is important in determining prognosis and helps to guide the management of patient.”
Can CT Image Reconstruction Parameters Impact the Predictive Value of Radiomics Features in Grading Pancreatic Neuroendocrine Neoplasms?
Tixier F, Lopez-Ramirez F, Blanco A, Yasrab M, Javed AA, Chu LC, Fishman EK, Kawamoto S.
Bioengineering (Basel). 2025 Jan 16;12(1):80. - “In this study, we assessed the impact of soft tissue image reconstruction kernels on the radiomics features, explored the possibility of correcting for this effect using ComBat harmonization, and evaluated the predictive value of the radiomic features from images reconstructed with B20f and I26f to distinguish between WHO grade 1 and higher grade PanNENs, including grade 2 or 3 PanNETs and PanNECs. The primary objective was to investigate the reconstruction variability to provide valuable insights to improve the generalizability of PanNENs grading models based on radiomics. However, the results on feature robustness to reconstruction kernel and ComBat harmonization should extend to other radiomic models based on contrast CT features obtained from images reconstructed with iterative or filtered back projection soft tissue reconstruction kernels.”
Can CT Image Reconstruction Parameters Impact the Predictive Value of Radiomics Features in Grading Pancreatic Neuroendocrine Neoplasms?
Tixier F, Lopez-Ramirez F, Blanco A, Yasrab M, Javed AA, Chu LC, Fishman EK, Kawamoto S.
Bioengineering (Basel). 2025 Jan 16;12(1):80.
- “A significant barrier to early detection even in high-risk cohorts is that the pancreas often appears morphologically normal during the pre-diagnostic phase. Yet, the disease can progress rapidly from subclinical stages to widespread metastasis, undermining the effectiveness of screening. Recently, artificial intelligence (AI) applied to cross-sectional imaging has shown significant potential in identifying subtle, early-stage changes in pancreatic tissue that are often imperceptible to the human eye. Moreover, AI-driven imaging also aids in the discovery of prognostic and predictive biomarkers, essential for personalized treatment planning. ”
AI-Driven insights in pancreatic cancer imaging: from pre-diagnostic detection to prognostication.
Antony A, Mukherjee S, Bi Y, Collisson EA, et al.
Abdom Radiol (NY). 2024 Dec 30. doi: 10.1007/s00261-024-04775-x. Epub ahead of print. PMID: 39738571. - ” Beyond early detection, AI plays a crucial role in prognostication and treatment response assessment for PDAC. Furthermore, AI-driven quantitative imaging can continuously monitor treatment response, offering dynamic insights into how PDAC tumors evolve during therapy. By analyzing changes in tumor size, shape, and internal heterogeneity, AI tools provide dynamic assessments that can guide treatment adjustments. This capability is particularly vital in PDAC, where conventional imaging often lags in detecting therapeutic impact, leading to delays in optimizing treatment strategies. ”
AI-Driven insights in pancreatic cancer imaging: from pre-diagnostic detection to prognostication.
Antony A, Mukherjee S, Bi Y, Collisson EA, et al.
Abdom Radiol (NY). 2024 Dec 30. doi: 10.1007/s00261-024-04775-x. Epub ahead of print. PMID: 39738571. - “Early imaging findings of PDAC are often subtle and easily overlooked on CT scans. Many healthy individuals may also show these indirect signs, as they are not specific to PDAC. Evaluating such findings can be subjective, with low inter-reader agreement. Frequently, the pancreas appears morphologically normal during the pre-diagnostic phase, resulting in missed diagnoses when the disease is still in its early, treatable stages.”
AI-Driven insights in pancreatic cancer imaging: from pre-diagnostic detection to prognostication.
Antony A, Mukherjee S, Bi Y, Collisson EA, et al.
Abdom Radiol (NY). 2024 Dec 30. doi: 10.1007/s00261-024-04775-x. Epub ahead of print. PMID: 39738571. - Frequently, the pancreas appears morphologically normal during the pre-diagnostic phase, resulting in missed diagnoses when the disease is still in its early, treatable stages. This highlights the critical necessity of imaging-based biomarkers to improve early detection of sporadic PDAC in high-risk groups. While the pancreas may seem normal to the human eye in these early stages, AI models, such as ML algorithms analyzing radiomic features (e.g., shape, intensity, texture) and DL models like convolutional neural networks (CNNs), excel at processing complex imaging data. These tools can identify subtle patterns that often escape traditional visual assessments.”
AI-Driven insights in pancreatic cancer imaging: from pre-diagnostic detection to prognostication.
Antony A, Mukherjee S, Bi Y, Collisson EA, et al.
Abdom Radiol (NY). 2024 Dec 30. doi: 10.1007/s00261-024-04775-x. Epub ahead of print. PMID: 39738571. - “AI-powered tools for pancreas segmentation enable extraction of critical biomarkers for early PDAC detection at the asymptomatic stage on pre-diagnostic CTs. The latter are defined as incidental CTs conducted for unrelated clinical indications between 3 and 36 months before a clinical PDAC diagnosis. They are typically interpreted as negative for PDAC during routine clinical evaluations and confirmed as such during data curation . It is crucial to distinguish pre-diagnostic CTs from diagnostic CTs where a mass is present but missed during routine interpretation. The latter are referred to as diagnostic CTs with missed cancer. ”
AI-Driven insights in pancreatic cancer imaging: from pre-diagnostic detection to prognostication.
Antony A, Mukherjee S, Bi Y, Collisson EA, et al.
Abdom Radiol (NY). 2024 Dec 30. doi: 10.1007/s00261-024-04775-x. Epub ahead of print. PMID: 39738571. - “The significant differences in radiomic signatures between scans with healthy pancreas and pre-diagnostic cases, acquired 6 months to 3 years before PDAC diagnosis with no visible tumors, were highlighted in a recent study . This analysis, conducted on an internal dataset of 66 patients (22 healthy controls, 22 pre-diagnostic, and 22 diagnostic CTs), identified specific radiomic features that clearly distinguished healthy from pre-diagnostic groups. A method using a step-by-step feature selection process and a simple classification model was developed, achieving a mean accuracy of 86% when categorizing scans as either healthy or pre-diagnostic, validated on an external dataset of 28 scans (14 in each group). Furthermore, the study also demonstrated incremental trends in these features as pre-diagnostic cases progressed toward diagnostic scans.”
AI-Driven insights in pancreatic cancer imaging: from pre-diagnostic detection to prognostication.
Antony A, Mukherjee S, Bi Y, Collisson EA, et al.
Abdom Radiol (NY). 2024 Dec 30. doi: 10.1007/s00261-024-04775-x. Epub ahead of print. PMID: 39738571. - “In addition to detection of subtle, pre-diagnostic imaging signatures within the pancreas, AI can also monitor systemic changes linked to early PDAC, a disease with well-known sequalae in many organ systems. These changes include variations in body composition and metabolic function that contribute to conditions like cancer-induced cachexia, which leads to muscle loss, with or without fat depletion . Cachexia can begin even in the preclinical stages of the disease, when precursor lesions are present . A recent study evaluated longitudinal changes in body composition, specifically, subcutaneous adipose tissue (SAT), visceral adipose tissue (VAT), skeletal muscle area (SMA), and vertebral bone area and density, to assess the potential of extra-pancreatic imaging markers for pre-diagnostic PDAC detection. ”
AI-Driven insights in pancreatic cancer imaging: from pre-diagnostic detection to prognostication.
Antony A, Mukherjee S, Bi Y, Collisson EA, et al.
Abdom Radiol (NY). 2024 Dec 30. doi: 10.1007/s00261-024-04775-x. Epub ahead of print. PMID: 39738571. - “Another DL model, PANDA (Pancreatic Cancer Detection with Artificial Intelligence), utilizes pancreas segmentation masks that include lesions as the initial step in a three-stage algorithm. The second stage focuses on lesion detection, identifying both PDAC and non-PDAC lesions, and is followed by differential diagnosis if a lesion is detected. The model was trained on a dataset of 3,208 non-contrast CT scans, where pancreas and lesion annotations were mapped from paired contrast-enhanced CT scans through image registration. PANDA achieved an AUC of 0.987 (95% CI 0.975–0.996) for PDAC identification on an internal dataset of 291 patients (108 PDAC scans, 67 non- PDAC lesion scans, and 116 control scans). On an external validation dataset of 5,337 patients (2,737 PDAC scans, 932 non-PDAC lesion scans, and 1,668 control scans), the model demonstrated a sensitivity of 90.1% (95% CI 89.0–91.2%) for PDAC identification. It maintained sensitivity above 90% for detecting stage 1 and stage 2 tumors across both datasets. ”
AI-Driven insights in pancreatic cancer imaging: from pre-diagnostic detection to prognostication.
Antony A, Mukherjee S, Bi Y, Collisson EA, et al.
Abdom Radiol (NY). 2024 Dec 30. doi: 10.1007/s00261-024-04775-x. Epub ahead of print. PMID: 39738571. - “Recent innovations aim to address the limitations of DL-based segmentation. A notable study developed a semi-automated, bounding-box CNN for PDAC segmentation, designed to focus on peri-tumoral regions rather than the entire pancreas. By limiting the input to this focused area, the CNN was able to significantly improve segmentation performance. This model was trained on the largest dataset reported to date—1,151 portal venous-phase CTs from treatment-naïve patients with biopsy-confirmed PDAC. The bounding-box CNN achieved a high DSC of 0.84 on the internal test subset and demonstrated excellent generalizability across two public datasets: Medical Segmentation Decathlon (MSD) (DSC: 0.82) and The Cancer Imaging Archive (TCIA) (DSC: 0.84) . The model’s ability to generalize across public datasets further validates its potential for clinical adoption. ”
AI-Driven insights in pancreatic cancer imaging: from pre-diagnostic detection to prognostication.
Antony A, Mukherjee S, Bi Y, Collisson EA, et al.
Abdom Radiol (NY). 2024 Dec 30. doi: 10.1007/s00261-024-04775-x. Epub ahead of print. PMID: 39738571. - ”Integrating multi-domain data from various sources, such as imaging and clinico-pathologic variables, has emerged as a powerful method for predicting patient outcomes. By merging diverse data sources, AI models can extract complex, multidimensional patterns, offering a more enriched dataset than unimodal approaches. For instance, recent studies that integrated radiomic features with clinical variables such as age, CA19-9 levels, tumor morphology, and metastatic status demonstrated superior performance in survival prediction compared to models relying on radiomics or clinical features alone .”
AI-Driven insights in pancreatic cancer imaging: from pre-diagnostic detection to prognostication.
Antony A, Mukherjee S, Bi Y, Collisson EA, et al.
Abdom Radiol (NY). 2024 Dec 30. doi: 10.1007/s00261-024-04775-x. Epub ahead of print. PMID: 39738571. - “Radiomic signatures have also been evaluated for their utility in predicting early recurrence in post-surgical patients. A recent model achieved an AUC of 0.73 in the training dataset and 0.67 in the validation dataset for predicting early recurrence. Multivariate analysis confirmed that radiomic signatures were independent predictors of early recurrence, supporting the robustness of AI-driven approaches in this context . Similarly, radiomics-based models have demonstrated high performance in predicting liver and lymph node metastasis, with AUCs of 0.7 and 0.85, respectively.”
AI-Driven insights in pancreatic cancer imaging: from pre-diagnostic detection to prognostication.
Antony A, Mukherjee S, Bi Y, Collisson EA, et al.
Abdom Radiol (NY). 2024 Dec 30. doi: 10.1007/s00261-024-04775-x. Epub ahead of print. PMID: 39738571. - “Publicly available datasets, vital for external validation, often lack quality and completeness, further limiting AI implementation in clinical practice. For instance, a recent study [66] assessing these datasets identified substantial quality gaps, including suboptimal imaging, incomplete annotations, and inherent biases. A significant proportion of CTs lacked essential details about the tumor histopathology, and approximately 25% include biliary stents. The presence of biliary stents introduces bias in AI models, as these models may erroneously associate the presence of stents with a PDAC diagnosis.”
AI-Driven insights in pancreatic cancer imaging: from pre-diagnostic detection to prognostication.
Antony A, Mukherjee S, Bi Y, Collisson EA, et al.
Abdom Radiol (NY). 2024 Dec 30. doi: 10.1007/s00261-024-04775-x. Epub ahead of print. PMID: 39738571. - “AI has shown great potential in transforming the early detection, diagnosis, and prognostication of PDAC, often outperforming radiologists by identifying subtle imaging patterns imperceptible to the human eye. This capability can improve patient outcomes by detecting PDAC at treatable stages. Beyond early detection, AI aids in prognostication by predicting patient outcomes, early recurrence, and metastasis risk with precision. Integrating non-pixel-based data can further enhance predictive accuracy, advancing precision medicine. However, challenges like limited pre-diagnostic data and model generalizability must be addressed for broader clinical adoption, requiring collaborative efforts to standardize data and expand high-quality datasets. As AI evolves, it is poised to enhance diagnostic accuracy, support personalized treatment planning, and improve survival rates, revolutionizing PDAC detection and treatment.”
AI-Driven insights in pancreatic cancer imaging: from pre-diagnostic detection to prognostication.
Antony A, Mukherjee S, Bi Y, Collisson EA, et al.
Abdom Radiol (NY). 2024 Dec 30. doi: 10.1007/s00261-024-04775-x. Epub ahead of print. PMID: 39738571.
Early detection of pancreatic cancer in the era of precision medicine.
Ahmed TM, Kawamoto S, Lopez-Ramirez F, Yasrab M, Hruban RH, Fishman EK, Chu LC.
Abdom Radiol (NY). 2024 Oct;49(10):3559-3573.- “While progress has been made in early detection strategies for PDAC, challenges persist. Screening protocols, liquid biopsy and AI techniques show promise in improving diagnoses, but their clinical adoption remains limited. Future integration of current divergent approaches into a fused multiomic framework remains necessary. Despite existing hurdles however, ongoing advances continue to make incremental progress, offering hope to enable early detection and to slowly but surely improve outcomes of this deadly disease.”
Early detection of pancreatic cancer in the era of precision medicine.
Ahmed TM, Kawamoto S, Lopez-Ramirez F, Yasrab M, Hruban RH, Fishman EK, Chu LC.
Abdom Radiol (NY). 2024 Oct;49(10):3559-3573. - “Another interesting avenue of AI developments that may be integrated into multifaceted frameworks are large language models (LLM). Medical documentation in electronic health records (EHR) is largely captured in unstructured text. LLMs such as GPT-4 (OpenAI) may therefore be uniquely suited to be applied to these datasets to streamline diagnostics and improve reporting. If securely integrated into EHRs, LLMs could swiftly mine EHRs to provide radiologists with clinically relevant information that may help frame imaging reports . An example of this could be to prompt LLMs to list all the risk factors a patient may have for pancreatic cancer and to stratify that patient’s risk based on an integration of the imaging findings with prior clinical features and demographics.”
Early detection of pancreatic cancer in the era of precision medicine.
Ahmed TM, Kawamoto S, Lopez-Ramirez F, Yasrab M, Hruban RH, Fishman EK, Chu LC.
Abdom Radiol (NY). 2024 Oct;49(10):3559-3573. 
- Our research concludes that the proposed CAD (computer-aided diagnosis) system for pancreatic cancer detection and classification using deep learning is effective, achieving good performance results. The proposed system aims to address the limitations of the manual identification of pancreatic tumors by radiologists, which is challenging and time-consuming due to the complex nature of CT scan images. The objective of the work is to apply a deep learning model to create a four-stage framework for the preprocessing, segmentation, detection, and classification of pancreatic cancers. The potential for this discovery to transform early pancreatic cancer detection and classification makes it significant. The suggested CAD system can greatly increase diagnostic efficiency and accuracy by automating the tumor identification and categorization process. This might result in early detection and the potential to save many lives worldwide.
Automated CAD system for early detection and classification of pancreatic cancer using deep learning model
Nadeem A et al.
PLOS ONE | https://doi.org/10.1371/journal.pone.0307900 January 3, 2025
Integrating OMICS-based platforms and analytical tools for diagnosis and management of pancreatic cancer: a review.
Sousa P, Silva L, Câmara JS, Guedes de Pinho P, Perestrelo R.
Mol Omics. 2024 Dec 23. doi: 10.1039/d4mo00187g.
Integrating OMICS-based platforms and analytical tools for diagnosis and management of pancreatic cancer: a review.
Sousa P, Silva L, Câmara JS, Guedes de Pinho P, Perestrelo R.
Mol Omics. 2024 Dec 23. doi: 10.1039/d4mo00187g.
Integrating OMICS-based platforms and analytical tools for diagnosis and management of pancreatic cancer: a review.
Sousa P, Silva L, Câmara JS, Guedes de Pinho P, Perestrelo R.
Mol Omics. 2024 Dec 23. doi: 10.1039/d4mo00187g.
Integrating OMICS-based platforms and analytical tools for diagnosis and management of pancreatic cancer: a review.
Sousa P, Silva L, Câmara JS, Guedes de Pinho P, Perestrelo R.
Mol Omics. 2024 Dec 23. doi: 10.1039/d4mo00187g.- To achieve the best possible outcomes for PaC patients, the development of new tests that could improve early diagnosis of the disease is essential. In this context, multi-OMICs approaches have emerged as a promising tool capable of revolutionizing both the early diagnosis and treatment of PaC. These integrative approaches not only have the potential to improve patient survival rates and quality of life, but also offer a personalized perspective for disease management. Researchers can develop more accurate predictive models for pre-diagnosis, and in depth analysis of molecular profiles allows the identification of patient subgroups with specific characteristics, paving the way for targeted and more effective and specific therapies.
Integrating OMICS-based platforms and analytical tools for diagnosis and management of pancreatic cancer: a review.
Sousa P, Silva L, Câmara JS, Guedes de Pinho P, Perestrelo R.
Mol Omics. 2024 Dec 23. doi: 10.1039/d4mo00187g. - “Similar improvements were observed when CA19-9 was combined with additional serum biomarkers, such as micro- RNAs. In a study of blood samples from 35 PaC patients and 15 CTLs, expression levels of selected microRNAs (miRNAs) and serum CA19-9 concentrations were determined by quantitative real-time reverse transcription-polymerase chain reaction (qRTPCR) and electrochemiluminescence immunoassay, respectively. Compared to CTLs, the levels of three miRNAs (miR-22- 3p, miR-642b-3p and miR-885-5p) were significantly higher in PaC patients, even in those with early-stage disease (IB and IIB). A panel of six miRNAs (let-7b-5p, miR-192-5p, miR-19a-3p, miR- 19b-3p, miR-223-3p, and miR-25-3p) together with serum miR- 25 in combination with CA19-9 and miR-17-5p methylation, showed superior diagnostic performance compared to CA19-9 or CEA.”
Integrating OMICS-based platforms and analytical tools for diagnosis and management of pancreatic cancer: a review.
Sousa P, Silva L, Câmara JS, Guedes de Pinho P, Perestrelo R.
Mol Omics. 2024 Dec 23. doi: 10.1039/d4mo00187g. - “There is also growing interest in the use of new approaches to detect potential biomarkers, highlighting the liquid biopsy, one of the most research hotspots that isolates the cancer-derived components from patients, including CTC, ctDNA, miRNA, lncRNA,99 which are present in body fluids such as blood, urine, and saliva. In addition, the development of modern technologies, including artificial intelligence (AI), can play a major role in the initial qualitative interpretation of cancer imaging, the prediction of clinical outcomes, and also the assessment of the impact of disease and treatment on adjacent organs. Particularly, the full integration of AI, molecular biomarkers, and complex intermolecular networks is expected to be a turning point in digital healthcare, ultimately improving personalized diagnostics and patient care.”
Integrating OMICS-based platforms and analytical tools for diagnosis and management of pancreatic cancer: a review.
Sousa P, Silva L, Câmara JS, Guedes de Pinho P, Perestrelo R.
Mol Omics. 2024 Dec 23. doi: 10.1039/d4mo00187g. - The synergy between AI, advanced biosensors and bioinformatics holds great promise for overcoming current limitations in cancer diagnosis. AI can identify subtle patterns and early biomarkers by analyzing large volumes of multi-OMICs data and advanced imaging, and continuously learning from clinical data to predict disease progression.101 Nanotechnology-based biosensors detect biomarkers at extremely low concentrations, such as ctDNA and miRNA, with high sensitivity and specificity, enabling rapid, accessible diagnostics and reducing reliance on invasive methods.102 By integrating and interpreting this complex data, bioinformatics can identify diagnostic clues and enable tailored treatments that improve efficacy and minimize side effects. This technological synergy has the potential to revolutionize oncology by enabling earlier detection, more precise treatment and improved quality of life for patients.
Integrating OMICS-based platforms and analytical tools for diagnosis and management of pancreatic cancer: a review.
Sousa P, Silva L, Câmara JS, Guedes de Pinho P, Perestrelo R.
Mol Omics. 2024 Dec 23. doi: 10.1039/d4mo00187g. - In parallel with these advances, analytical tools have influenced the study and management of PaC. MS, including high-resolution LC–MS/MS, MALDI Imaging MS, and GC–MS, allows in-depth profiling of proteins, metabolites, and their spatial changes in tumour tissue. Next-generation sequencing, including whole genome/exome sequencing and single-cell RNA sequencing, reveals the genetic landscape and tumour heterogeneity. Advanced imaging techniques, such as multiplex immunohistochemistry and fluorescent in situ hybridization, provide insight into cellular interactions and specific genetic alterations. Integrating this data with sophisticated bioinformatics approaches enables the identification of more precise biomarkers for early detection and the development of personalised therapies. As a result, these tools are significantly improving our understanding of PaC biology, potentially leading to more accurate diagnosis and improved patient outcomes.
Integrating OMICS-based platforms and analytical tools for diagnosis and management of pancreatic cancer: a review.
Sousa P, Silva L, Câmara JS, Guedes de Pinho P, Perestrelo R.
Mol Omics. 2024 Dec 23. doi: 10.1039/d4mo00187g.
- ”This review provides a comprehensive overview of the current landscape in pancreatic cancer (PC) screening, diagnosis, and early detection. This emphasizes the need for targeted screening in high-risk groups, particularly those with familial predispositions and genetic mutations, such as BRCA1, BRCA2, and PALB2. This review highlights the sporadic nature of most PC cases and significant risk factors, including smoking, alcohol consumption, obesity, and diabetes. Advanced imaging techniques, such as Endoscopic Ultrasound (EUS) and Contrast-Enhanced Harmonic Imaging (CEH-EUS), have been discussed for their superior sensitivity in early detection. This review also explores the potential of novel biomarkers, including those found in body fluids, such as serum, plasma, urine, and bile, as well as the emerging role of liquid biopsy technologies in analyzing circulating tumor DNA (ctDNA), circulating tumor cells (CTCs), and exosomes. AI-driven approaches, such as those employed in Project Felix and CancerSEEK, have been highlighted for their potential to enhance early detection through deep learning and biomarker discovery.”
Informatics strategies for early detection and risk mitigation in pancreatic cancer patients.
Jin D, Khan NU, Gu W, Lei H, Goel A, Chen T.
Neoplasia. 2025 Jan 21;60:101129. doi: 10.1016/j.neo.2025.101129. - ”Pancreatic ductal adenocarcinoma (PDAC), which constitutes approximately 90% of pancreatic cancer cases, is particularly notorious for its late-stage presentation. The lack of early symptoms and the deep anatomical location of the pancreas contribute to the challenge of timely diagnosis. Risk factors include smoking, obesity, diabetes mellitus, and a family history of the disease, with smoking alone doubling this risk. Notably, new-onset diabetes in older individuals is both a symptom and risk factor for pancreatic cancer, highlighting the need for vigilant monitoring. The early detection of pancreatic cancer remains a significant hurdle due to the absence of standardized screening protocols and reliable biomarkers. ”
Informatics strategies for early detection and risk mitigation in pancreatic cancer patients.
Jin D, Khan NU, Gu W, Lei H, Goel A, Chen T.
Neoplasia. 2025 Jan 21;60:101129. doi: 10.1016/j.neo.2025.101129. - “Precursor lesions, including pancreatic intraepithelial neoplasia (PanIN), intraductal papillary mucinous neoplasms (IPMNs), and mucinous cystic neoplasms (MCNs), are often detected by imaging, but only a tiny fraction progress to high-grade neoplasia. Molecular evidence indicates that most PDACs originate from PanINs, which are not visible with current imaging techniques, thus complicating early detection. The extended timeline of progression from precursors to invasive PDAC further challenges timely intervention and effective management. ”
Informatics strategies for early detection and risk mitigation in pancreatic cancer patients.
Jin D, Khan NU, Gu W, Lei H, Goel A, Chen T.
Neoplasia. 2025 Jan 21;60:101129. doi: 10.1016/j.neo.2025.101129. - “The lifetime risk of pancreatic ductal adenocarcinoma (PDAC) is approximately 1.7% in an average individual, but several clinical factors can significantly increase this risk. Chronic pancreatitis, mainly hereditary or recurrent, contributes to PDAC risk through ongoing tissue injury, inflammation, and DNA damage. However, recent estimates suggest a risk reduction to below 10% from earlier estimates as high as 70%. Acute pancreatitis, on the other hand, is linked to a heightened risk of PDAC within one year of the episode, likely due to tumor-induced obstruction. Smoking doubles the risk of PDAC, with meta-analyses indicating that smoking accounts for 11%–32% of pancreatic cancer cases. Heavy alcohol consumption also modestly increases PDAC risk, whereas light drinking does not. Obesity is associated with an elevated risk of PDAC, with studies showing a relative risk of 1.72 for individuals with a body mass index (BMI) of over 30. ”
Informatics strategies for early detection and risk mitigation in pancreatic cancer patients.
Jin D, Khan NU, Gu W, Lei H, Goel A, Chen T.
Neoplasia. 2025 Jan 21;60:101129. doi: 10.1016/j.neo.2025.101129. - “Intraductal papillary mucinous neoplasms (IPMNs) and mucinous cystic neoplasms (MCNs) of the pancreas offer opportunities for early detection of premalignant lesions, potentially targeting early detection strategies. However, their general applicability to pancreatic ductal adenocarcinoma (PDAC) is limited because premalignant cystic lesions serve as precursors for PDAC in only approximately 15% of cases. Despite this, IPMNs, particularly those in the main pancreatic duct, may harbor PDACs, thus warranting specialized surveillance or surgical resection based on lesion size and morphology. Additionally, individuals undergoing surveillance for IPMNs who later develop PDACs separate from IPMNs are not uncommon, given that PanINs, often undetectable by imaging, are generally more abundant in resected pancreata than in IPMNs.”
Informatics strategies for early detection and risk mitigation in pancreatic cancer patients.
Jin D, Khan NU, Gu W, Lei H, Goel A, Chen T.
Neoplasia. 2025 Jan 21;60:101129. doi: 10.1016/j.neo.2025.101129. - “Some current examples of AI technology in this area include Project Felix at Johns Hopkins University. Researchers developed a deep learning system for detecting small pancreatic tumors using CT scan images. Project Felix has brought significant improvement in the detection of pancreatic cancer, especially the first and second stages, and this has been tested in a real-world setting through clinical trials among patients with pancreatic cancer. CancerSEEK , a microfluidic platform coated with nanoparticles for identifying genetic mutation and protein biomarkers, using AI for biomarker identification, can detect several types of cancer, including pancreatic cancer. The earlier researcher showed that through CancerSEEK PDACs could be diagnosed, especially at early stages, with high sensitivity and specificity.”
Informatics strategies for early detection and risk mitigation in pancreatic cancer patients.
Jin D, Khan NU, Gu W, Lei H, Goel A, Chen T.
Neoplasia. 2025 Jan 21;60:101129. doi: 10.1016/j.neo.2025.101129. - “Personalized medicine is another significant focus, where genomic and proteomic data are integrated to tailor diagnostic and therapeutic approaches for individual patients. This involves leveraging patient- specific data to predict disease risk, optimize treatment plans, and improve outcomes. Personalized diagnostic tools are being developed to match patients with targeted therapies based on their unique genetic profiles and tumor characteristics [134]. Furthermore, ongoing research on pancreatic cancer’s tumor microenvironment and immune landscape is expected to yield new diagnostic and therapeutic strategies. Understanding the interactions between pancreatic cancer cells and their surrounding stroma and the role of immune cells can lead to the development of novel biomarkers and targeted therapies that address the complexities of the disease [135]. The future of pancreatic cancer diagnosis lies in the convergence of advanced imaging technologies, molecular and genetic profiling, AI-driven analysis, and personalized approaches. These advancements are poised to improve early detection, enhance diagnostic accuracy, and ultimately lead to improved patient outcomes.”
Informatics strategies for early detection and risk mitigation in pancreatic cancer patients.
Jin D, Khan NU, Gu W, Lei H, Goel A, Chen T.
Neoplasia. 2025 Jan 21;60:101129. doi: 10.1016/j.neo.2025.101129. - “AI-driven tools such as Project Felix and CancerSEEK offer new possibilities for enhancing diagnostic accuracy, refining risk assessment, and personalizing treatment. This strategic application of informatics in pancreatic cancer represents a comprehensive approach to improve early detection and risk management. Continued research and innovation are crucial for overcoming the existing limitations and advancing the fight against this formidable disease, ultimately aiming to save lives and improve patient outcomes.”
Informatics strategies for early detection and risk mitigation in pancreatic cancer patients.
Jin D, Khan NU, Gu W, Lei H, Goel A, Chen T.
Neoplasia. 2025 Jan 21;60:101129. doi: 10.1016/j.neo.2025.101129.
Informatics strategies for early detection and risk mitigation in pancreatic cancer patients.
Jin D, Khan NU, Gu W, Lei H, Goel A, Chen T.
Neoplasia. 2025 Jan 21;60:101129. doi: 10.1016/j.neo.2025.101129.
Informatics strategies for early detection and risk mitigation in pancreatic cancer patients.
Jin D, Khan NU, Gu W, Lei H, Goel A, Chen T.
Neoplasia. 2025 Jan 21;60:101129. doi: 10.1016/j.neo.2025.101129.- • Radiology plays an important role in the initial diagnosis and staging of pancreatic cancer.
• CT is the preferred modality over MRI due to wider availability, greater consistency in image quality, and lower cost.
• Patients can be triaged into resectable, borderline resectable, and locally advanced based on tumor involvement of arteries and veins.
• Accuracy of diagnosis and staging critically depends on the imaging technique and experience of the radiologists.
• Artificial intelligence has the potential to function as ‘second readers’ to improve the detection of small early stage tumors and provide imaging biomarkers to predict patient prognosis. .
Pancreatic ductal adenocarcinoma staging: a narrative review of radiologic techniques and advances.
Chu LC, Fishman EK
Int J Surg. 2024 Oct 1;110(10):6052-6063. - PDACs classically present as hypoenhancing masses with associated pancreatic duct dilatation and glandular atrophy of the body and tail. Pancreatic head tumors can cause common bile duct dilatation in addition to pancreatic duct dilatation, also known as the ‘double duct sign’. Up to 20% of PDACs enhance to the same degree as the background pancreas, and this isoattenuating pattern is more commonly found with smaller ( ≤20 mm) tumors. These small isoattenuating tumors can be difficult to detect on CT; therefore, radiologists often rely on secondary signs of the pancreatic duct or common bile duct dilatation for tumor detection. MRI and PET/CT have reported sensitivities of 79.2 and 73.7% in the detection of isoattenuating tumors, respectively, and may aid in detecting suspected pancreatic tumors that are occult on CT.”
Pancreatic ductal adenocarcinoma staging: a narrative review of radiologic techniques and advances.
Chu LC, Fishman EK
Int J Surg. 2024 Oct 1;110(10):6052-6063. - “Both arterial and venous involvement are pivotal in determining resectability. Based on the National Comprehensive Cancer Network (NCCN) guidelines, tumors without arterial tumor contact or superior mesenteric vein (SMV) or portal vein (PV) tumor contact are considered resectable. Tumors with ≤ 180° contact with the SMV or PV without contour irregularity are also considered resectable. Arterial abutment of the celiac artery or superior mesenteric artery (SMA) (< 180°) is considered borderline resectable, whereas arterial encasement ( ≥ 180° ) is usually considered locally advanced.”
Pancreatic ductal adenocarcinoma staging: a narrative review of radiologic techniques and advances.
Chu LC, Fishman EK
Int J Surg. 2024 Oct 1;110(10):6052-6063. - “The reported accuracy in determining tumor resectability ranges from 73 to 87% for CT and 70 to 79% for MRI, although this may depend on radiologists’ experience. CT offers superior spatial resolution and is less susceptible to artifacts compared to MRI. Also, CT allows for greater confidence in the assessment of tumor–vascular relationships. MRI is a critical problem-solving tool in the characterization of indeterminate liver lesions, which influences staging localized vs. metastatic disease. PET lacks the spatial resolution critical for the staging of locoregional involvement and is not used routinely in staging.”
Pancreatic ductal adenocarcinoma staging: a narrative review of radiologic techniques and advances.
Chu LC, Fishman EK
Int J Surg. 2024 Oct 1;110(10):6052-6063.
Pancreatic ductal adenocarcinoma staging: a narrative review of radiologic techniques and advances.
Chu LC, Fishman EK
Int J Surg. 2024 Oct 1;110(10):6052-6063.
Pancreatic ductal adenocarcinoma staging: a narrative review of radiologic techniques and advances.
Chu LC, Fishman EK
Int J Surg. 2024 Oct 1;110(10):6052-6063.- “Cinematic rendering can accentuate subtle texture changes and improve tumor conspicuity relative to traditional 2D images, 3D volume rendering, or maximum intensity projection images. Cinematic rendering may be able to enhance the visualization of spatial relationships among the tumor and adjacent vasculature, differentiating true tumor infiltration from simple proximity to vessels. This can potentially improve the assessment of resectability and assist in determining optimal vascular reconstruction options. Cinematic rendering vascular maps illustrate the major arteries and veins with exquisite detail and can highlight the presence of variant vascular anatomy that may increase the risk of complications, such as hemorrhage, ischemia, anastomotic leakage, or pseudoaneurysm formation.”
Pancreatic ductal adenocarcinoma staging: a narrative review of radiologic techniques and advances.
Chu LC, Fishman EK
Int J Surg. 2024 Oct 1;110(10):6052-6063. - “Radiomics converts imaging data into high-dimensional features that can be used to characterize spatial heterogeneity inherent in disease processes. The features of radiomics can be classified into signal intensity, shape, and texture. Signal intensity (first-order) features are derived from histograms of individual voxel signal intensities, providing measures of central tendency and shape of the distribution. Shape features are extracted from the three-dimensional surface of the region of interest. Texture features are calculated in three dimensions, considering the correlation of signal intensities of adjacent voxels. In addition, feature extraction may be performed after applying a secondary filter, such as a wavelet or Gaussian filter.”
Pancreatic ductal adenocarcinoma staging: a narrative review of radiologic techniques and advances.
Chu LC, Fishman EK
Int J Surg. 2024 Oct 1;110(10):6052-6063. - “AI can theoretically function as ‘second readers’ to improve radiologists’ sensitivity in the detection of small tumors, which potentially can be cured with surgical resection. A preliminary study by Liu et al. showed promising results suggesting that DL could accurately differentiate CT scans of patients with PDAC from CT scans of healthy controls. More recently, Chen et al. developed a DL tool that differentiated CT scans of patients with PDAC vs. healthy controls with 89.9% sensitivity, 95.9% specificity, and 93.4% accuracy in the local test set. They validated this DL tool on a Taiwanese nationwide external validation set and achieved 89.7% sensitivity, 92.8% specificity, and 91.4% accuracy. Also, Park et al. developed a different DL tool that achieved high sensitivity comparable to radiologists in the detection of not only pancreatic solid masses (98–100%) but also cystic masses 1.0 cm or larger (sensitivity 92–93%), bringing us closer to a universal pancreatic neoplasm detector.”
Pancreatic ductal adenocarcinoma staging: a narrative review of radiologic techniques and advances.
Chu LC, Fishman EK
Int J Surg. 2024 Oct 1;110(10):6052-6063. - “Preliminary studies using emerging technologies such as advanced visualization and AI have revealed the potential of these tools to improve the initial diagnosis and staging of patients with PDAC. However, there remain several limitations. Most of these studies have been single-center retrospective studies, and their promising results should be validated in future multicenter prospective studies. Secondly, one of the major criticisms of AI is its ‘blackbox’ nature, making it difficult for clinicians to decipher the rationale behind AI predictions. Explainable or ‘glassbox’ AI is an active area of research that aims to render AI models more easily understandable and may help improve their clinical acceptance. Thirdly, these tools should be integrated seamlessly into the workflow to ensure widespread clinical implementation.”
Pancreatic ductal adenocarcinoma staging: a narrative review of radiologic techniques and advances.
Chu LC, Fishman EK
Int J Surg. 2024 Oct 1;110(10):6052-6063. - “Radiology plays a significant role in the initial diagnosis and staging of patients with PDAC, triaging patients with resectable disease, and determining treatment response to neoadjuvant chemotherapy and radiation. CT is the most used radiologic modality for PDAC staging, with MRI and PET/CT usually reserved as problem-solving tools. Current challenges in staging include preoperative diagnosis of lymph node metastases, subtle liver and peritoneal metastases, and R0 resection following neoadjuvant therapy. Artificial intelligence offers the potential of earlier disease diagnosis at the localized disease stage and prognostic radiologic biomarkers to optimize patient management, which can help improve patient outcomes.”
Pancreatic ductal adenocarcinoma staging: a narrative review of radiologic techniques and advances.
Chu LC, Fishman EK
Int J Surg. 2024 Oct 1;110(10):6052-6063.
Early detection of pancreatic cancer in the era of precision medicine.
Ahmed TM, Kawamoto S, Lopez-Ramirez F, Yasrab M, Hruban RH, Fishman EK, Chu LC.
Abdom Radiol (NY). 2024 Oct;49(10):3559-3573. doi: 10.1007/s00261-024-04358-w. Epub 2024 May 18. PMID: 38761272.- “While the 5-year survival rate for patients with localized PDAC is a favorable 44%, the aggregated overall survival rate across all stages drops to 13% . This negative skew, and a major driver of the particularly poor outcomes of this disease, is because only 9% of patients present with localized disease at the time of diagnosis. Consequently, only a small fraction (10–20%) of patients may be eligible for surgical resection, which remains the only curative treatment option. Although earlier detection of disease is challenging given the nonspecific presentation and asymptomatic course of early PDAC, prior studies have demonstrated that the median survival of incidentally detected PDACs is 10 months greater than that of comparable symptomatic PDAC. Smaller tumors are also more amenable to surgical resection with resectability rates of up to 99% in tumors under 2 cm in size at the time of detection. Early detection therefore holds promise to significantly improve outcomes of pancreatic cancer.”
Early detection of pancreatic cancer in the era of precision medicine.
Ahmed TM, Kawamoto S, Lopez-Ramirez F, Yasrab M, Hruban RH, Fishman EK, Chu LC.
Abdom Radiol (NY). 2024 Oct;49(10):3559-3573. - “Amongst identified high-risk conditions, the International Cancer of the Pancreas Surveillance (CAPS) consortium consensus statement recommends screening for all patients with Peutz Jeghers syndrome (deleterious germline STK11 variants) and familial atypical multiple mole melanoma syndrome (deleterious germline CDKN2A variants) starting at 40 years of age. Carriers of deleterious variants in BRCA2, BRCA1, PALB2, ATM, MLH1, MSH2 or MSH6 with at least one affected first-degree blood relative are recommended to begin screening between 45 and 50 years of age It is estimated that these inherited genetic cancer syndromes account for 3–5% of cases of PDAC . Individuals with familial pancreatic cancer, defend as having two or more first degree relatives with pancreatic cancer, are also recommended to undergo screening. Patients with familial pancreatic cancer are estimated to constitute 5–10% of cases of PDAC.”
Early detection of pancreatic cancer in the era of precision medicine.
Ahmed TM, Kawamoto S, Lopez-Ramirez F, Yasrab M, Hruban RH, Fishman EK, Chu LC.
Abdom Radiol (NY). 2024 Oct;49(10):3559-3573. - ‘Liquid biopsy’, which refers to the sampling of biomarkers from bodily fluids, has garnered increased attention in recent years. Compared to traditional biopsies, liquid biopsies are minimally invasive, allow for convenient serial sampling and may better capture tumor heterogeneity—rendering them ideal screening tools. Currently, targets of liquid biopsy primarily include circulating tumor DNA (ctDNA), circulating tumor cells (CTCs), and exosomes. While CTCs and exosomes have been an exciting area of new research, ctDNA remains the most widely studied targets for liquid biopsy and hold greatest clinical utility.”
Early detection of pancreatic cancer in the era of precision medicine.
Ahmed TM, Kawamoto S, Lopez-Ramirez F, Yasrab M, Hruban RH, Fishman EK, Chu LC.
Abdom Radiol (NY). 2024 Oct;49(10):3559-3573. - “The KRAS gene has been demonstrated to be the most common somatically mutated gene in PDAC. Approximately 95% of PDAC harbors mutant KRAS alleles, and tests that can detect these mutations have demonstrated impressive sensitivity for early-stage detection, with a previous study demonstrating that up to a third of patients with low-stage PDAC had detectable KRAS or GNAS mutations in circulating DNA. A number of studies have also evaluated the combined performance of ctDNA with other biomarkers . Notably, Cohen et al. demonstrated that KRAS ctDNA detection combined with a protein biomarker assay consisting of CA19-9, CEA, hepatocyte growth factor (HGF) and osteopontin (OPN) increased the sensitivity compared to detecting KRAS alone in suggesting the diagnosis of PDAC .”
Early detection of pancreatic cancer in the era of precision medicine.
Ahmed TM, Kawamoto S, Lopez-Ramirez F, Yasrab M, Hruban RH, Fishman EK, Chu LC.
Abdom Radiol (NY). 2024 Oct;49(10):3559-3573. - “Emerging studies have also ventured beyond application of AI techniques as ‘second readers’ on diagnostic scans and investigated the ability of AI models to detect visually occult PDAC on prediagnostic images months before the cancer is discernable to the human eye. Korfatis et al. demonstrated that a deep learning model could classify PDAC and controls on CT images with AUC of 0.91 at a median lead time of 475 days before clinical diagnosis. Other similar studies on this have reported an accuracy of 89% to 100% at a lead time ranging from 3 to 36 months before clinical diagnosis. Augmentation of current screening protocols with such models that specialize in detecting preinvasive lesions, particularly in conjunction with blood-based biomarkers holds potential to further streamline noninvasive screening strategies.”
Early detection of pancreatic cancer in the era of precision medicine.
Ahmed TM, Kawamoto S, Lopez-Ramirez F, Yasrab M, Hruban RH, Fishman EK, Chu LC.
Abdom Radiol (NY). 2024 Oct;49(10):3559-3573. - “Perhaps the most significant challenge of screening for PDAC is that of balancing the sensitivity of screening programs with the considerable risks of pancreatic surgery. This is particularly pressing as up to nearly 70% of surgeries undergone by HRIs under surveillance for PDAC, were, in retrospect, unnecessary. IPMNs contribute to the complexity of this challenge with prior studies demonstrating that only 25% of branch duct IPMN and 66% of main-duct IPMN that are surgically resected harbor high-grade dysplasia or an associated invasive carcinoma. Enhanced preoperative evaluation of IPMN dysplasia is therefore crucial to avoid unnecessary surgeries and optimize current screening protocols. “
Early detection of pancreatic cancer in the era of precision medicine.
Ahmed TM, Kawamoto S, Lopez-Ramirez F, Yasrab M, Hruban RH, Fishman EK, Chu LC.
Abdom Radiol (NY). 2024 Oct;49(10):3559-3573.
Vascular
- The classic radiologic findings were first described by Edward B. Singleton and David Merten in 1973. Typical radiographic appearances include skeletal demineralization, expanded shafts of the metacarpals and phalanges with widened medullary cavities, cardiomegaly, and intramural calcification of the proximal aorta with occasional extension into the aortic or mitral valves. Other commonly seen radiographic findings include shallow acetabular fossa, subluxation of the femoral head, coxa valga, hypoplastic radial epiphysis, soft tissue calcifications between the radius and ulna, constriction of the proximal radial shaft, acro-osteolysis, and equinovarus foot deformities.
- “The risk of aortic dissection and rupture is directly correlated with aortic diameter, with a sharp increase in risk when the ascending aorta measures 6.0 cm or the descending aorta measures 7.0 cm. At this size, there is a 14.1% risk of rupture, dissection, or death per year. For this reason, intervention is recommended when the ascending aorta reaches 5.5 cm or the descending aorta reaches 6.5 cm. For cases of familial aortic aneurysm or connective tissue disease, the threshold is even lower.”
Imaging of the Postoperative Thoracic Aorta
William Truesdell, Smita Pate
Radiol Clin N Am - (2024) in press - “Supracoronary ascending aortic replacement (SCAAR) involves resection of the ascending aorta distal to the sinuses of Valsalva and replacement with an interposition graft. When combined with concurrent replacement of the aortic valve, this is termed the Wheat procedure. This technique can be used to repair the ascending aorta when the underlying pathology spares the aortic annulus and root. SCAAR leaves the native coronary anatomy in place, avoiding potential complications associated with coronary buttons, and has the additional benefit of retaining the flow dynamics of the native coronary ostia. A drawback of the SCAAR is that there may be continued dilation of the aortic root, necessitating reoperation with root replacement.”
Imaging of the Postoperative Thoracic Aorta
William Truesdell, Smita Pate
Radiol Clin N Am - (2024) in press - “One of the most feared complications of endovascular stent grafts is spinal cord ischemia. This can occur from occlusion of the spinal arteries, and is at the greatest risk if the stent graft extends inferiorly below the T8 vertebral body, likely due to coverage of the artery of Adamkiewicz. Interestingly, this complication is less common after treatment of chronic type B aortic dissections, likely due to existing collateral circulation of the spinal cord. This is not directly diagnosed on postoperative CT surveillance; however it can be suggested in symptomatic patients with stent graft coverage of the related spinal arteries.”
Imaging of the Postoperative Thoracic Aorta
William Truesdell, Smita Pate
Radiol Clin N Am - (2024) in press - Singleton-Merten syndrome (SGMRT) is an uncommon autosomal dominant disorder characterized by abnormalities of blood vessels, teeth, and bone. Calcifications of the aorta and aortic and mitral valves occur in childhood or puberty and can lead to early death. Dental findings include delayed primary tooth exfoliation and permanent tooth eruption, truncated tooth root formation, early-onset periodontal disease, and severe root and alveolar bone resorption associated with dysregulated mineralization, leading to tooth loss. Osseous features consist of osteoporosis, either generalized or limited to distal extremities, distal limb osteolysis, widened medullary cavities, and easy tearing of tendons from bone. Less common features are mild facial dysmorphism (high anterior hair line, broad forehead, smooth philtrum, thin upper vermilion border), generalized muscle weakness, psoriasis, early-onset glaucoma, and recurrent infections. The disorder manifests with variable inter- and intrafamilial phenotypes (summary by Rutsch et al., 2015). Genetic Heterogeneity of Singleton-Merten Syndrome An atypical form of Singleton-Merten syndrome (SGMRT2; 616298) is caused by mutation in the DDX58 gene (609631) on chromosome 9p21.
- Upper limits of normal aortic diameter by segment: aortic root 4.0 cm, ascending aorta 4.0 cm or less than 1.5 times the descending aortic diameter, aortic arch 3.5 cm, descending aorta 3.0 cm. The new classification for aortic dissection from 2020 describes the entry tear zone, which determines the dissection type A/B, followed by the subscripts that denote the proximal and distal extensions according to the involved aortic zones. Although penetrating aortic ulcer (PAU) is categorized as part of the acute aortic syndrome , the vast majority of PAUs are asymptomatic. Symptomatic PAU is often associated with aortic wall hemorrhage. Ascending aortic dilation ranges from 4.0 to 4.4 cm, with an aneurysm defined at 4.5 cm.
Preoperative Imaging of the Thoracic Aorta
Zehavit E. Kirshenboim et al.
Radiol Clin N Am - (2024) in press - Aortic diameters vary by age and gender. For simplicity, the following upper limits of normal can be used: aortic root 4.0 cm, ascending aorta 4.0 cm or less than 1.5 times the descending aortic diameter, aortic arch 3.5 cm, descending aorta 3.0 cm. The normal ascending aortic wall thickness is approximately 2 mm.
Preoperative Imaging of the Thoracic Aorta
Zehavit E. Kirshenboim et al.
Radiol Clin N Am - (2024) in press - The aorta dilates with aging at a rate of approximately 0.7 mm per decade in women and 0.9 mm per decade in men. Thickening of the aortic wall and degeneration of the collagen and elastic components lead to aortic dilation, elongation, and tortuosity with advancing age, a process known as arteriosclerosis. Repeated pulsatile stress causes fragmentation of the elastic components in the proximal aorta, replacement by fibrotic tissue, and resultant aortic wall stiffening and increased pulse pressure, increasing the left ventricular workload.
Preoperative Imaging of the Thoracic Aorta
Zehavit E. Kirshenboim et al.
Radiol Clin N Am - (2024) in press - Acute IMH is contained hemorrhage within the aortic media, which can be attributed to PAU, trauma, or thrombosed dissected false lumen with microscopic tears in the intima. IMH accounts for 5% to 15% of AAS cases and occurs more frequently in older patients (60–80 years of age) compared to those with aortic dissection. Other risk factors include hypertension and aortic dilation, with commonly found concomitant abdominal aortic aneurysm. IMH is classified similarly to aortic dissection, with most cases (58%) being a type B.
Preoperative Imaging of the Thoracic Aorta
Zehavit E. Kirshenboim et al.
Radiol Clin N Am - (2024) in press - The two most common forms of inflammatory (noninfectious) aortitis are Takayasu arteritis and giant cell arteritis (GCA). Takayasu (necrotizing) arteritis typically affects females under 40 year old and usually manifests as panaortitis with granulomatous inflammation which may cause stenosis of the inflamed vessels, particularly the arch branches. GCA is more common in patients over 50 year old, is associated with temporal arteritis and polymyalgia rheumatica, syphilis (known as luetic aortopathy), mycobacterium tuberculosis infection, and human immunodeficiency virus. Contrast-enhanced CT is typically the preferred imaging modality for diagnosis, with key radiological findings including aortic wall thickening, periaortic fluid or soft tissue, rapid development of saccular aneurysms, and occasionally, the presence of air within the aortic wall.The term “mycotic aneurysm” refers to aneurysms caused by an infection, which are characterized by a mushroom-shaped appearance, and does not indicate fungal infection.
Preoperative Imaging of the Thoracic Aorta
Zehavit E. Kirshenboim et al.
Radiol Clin N Am - (2024) in press