Imaging Pearls ❯ June 2025
|
-- OR -- |
|
3D and Workflow
- Many of the lessons from venture capital can be applied to radiology, beginning with the inclusion of diverse skill sets designed to address historical gender inequity. Although many other specialties have gradually or even completely equalized in their gender distribution, since 2000, the representation of women in academic radiology has only marginally improved. Part of the issue is attributed to the “leaky pipeline” in radiology for Women have affected gender representation at every career stage. Residents, junior faculty members, Senior faculty members and leaders have left the field for various reasons, including juggling work and home responsibilities, fewer opportunities, and other more personal reasons. Intentional strategies by men and women in leadership positions to create more opportunities for women in radiology have the potential to benefit medicine, just as they benefit entrepreneurship.
Data-Driven Ventures: A Path to Higher Returns and Greater Gender Equity.
Abramson J, Fishman EK, Chu LC, Rowe SP, Crawford CK.
J Am Coll Radiol. 2025 Apr 24:S1546-1440 - “All careers, especially medical ones, are built on simultaneous individual and organizational investment of time and resources to train future practicing professionals. Similar to business, radiology should have a positive return on investment, which will be the direct result of investing in women’s initiatives for radiology by reduce or even halt the detrimental impact of the leaky pipeline. In addition, diversifying medical perspectives, regardless of gender, allows a more holistic approach to patient care, challenging one-sided diagnoses, and ensuring that all aspects of a patient’s condition are considered, resulting in better quality of care.”
Data-Driven Ventures: A Path to Higher Returns and Greater Gender Equity.
Abramson J, Fishman EK, Chu LC, Rowe SP, Crawford CK.
J Am Coll Radiol. 2025 Apr 24:S1546-1440
Adrenal
- “Tumors ≥ 4 cm, or those with indeterminate features may require further imaging, such as magnetic resonance imaging (MRI) or positron emission tomography (PET)/CT. The management of adrenal incidentalomas is determined by hormonal secretion and imaging characteristics. Surgical resection is recommended for functional tumors and those that are suspicious for malignancy, including tumors ≥ 4 cm in size and those with rapid growth. Non-functional tumors < 4 cm may undergo imaging surveillance.”
The Landmark Series: Evaluation and Management of Adrenal Incidentalomas.
Owei L, Wachtel H.
Ann Surg Oncol. 2025 Apr 30. doi: 10.1245/s10434-025-17296-8. Epub ahead of print. PMID: 40304946. - “A more recent review of 14 studies found that adenomas comprised 80–85% of adrenal incidentalomas. Of the adenomas, 40–70% were non-functioning, 20–50% were associated with mild autonomous cortisol secretion (MACS), 1–4% had overt Cushing’s syndrome, 2–5% were aldosterone secreting, and 1–5% were pheochromocytomas. Despite the large variability between studies, the general trend is that non-functional adenomas are the most frequent diagnosis, followed by cortisol-secreting, aldosterone-secreting, and catecholamine-secreting tumors respectively, while primary adrenocortical carcinoma (ACC) remains rare.”
The Landmark Series: Evaluation and Management of Adrenal Incidentalomas.
Owei L, Wachtel H.
Ann Surg Oncol. 2025 Apr 30. doi: 10.1245/s10434-025-17296-8. Epub ahead of print. PMID: 40304946. - “Non-contrast, or unenhanced, CT is the preferred initial imaging modality for the evaluation of adrenal incidentalomas .CT allows assessment of tissue density or tissue attenuation and is essential to assess the likelihood of malignancy in adrenal lesions. An adrenal mass is considered benign if it appears homogeneous and lipid-rich, with a density of ≤10 HU on an unenhanced CT scan. While standard contrast-enhanced CT is not suitable for distinguishing benign from malignant adrenal tumors, CT with delayed washout is more helpful. An adrenal protocol CT includes an unenhanced sequence followed by intravenous contrast administration with repeat imaging at 60–75 s (venous phase) and again at 15 min (delayed phase). Injected intravenous contrast washes out of benign lesions more rapidly than malignant lesions. Specifically, benign adenomas demonstrate an absolute percentage washout >60% at 15 min delay, which is therefore used as a key diagnostic cut-off.”
The Landmark Series: Evaluation and Management of Adrenal Incidentalomas.
Owei L, Wachtel H.
Ann Surg Oncol. 2025 Apr 30. doi: 10.1245/s10434-025-17296-8. Epub ahead of print. PMID: 40304946. - Tumor size is a key characteristic that determines management, due to the correlation between tumor size and risk of malignancy. In 2002, the National Institutes of Health (NIH) published a consensus recommendation on adrenal incidentalomas, which classified adrenal tumors measuring <4 cm as having a low risk of malignancy, those between 4 and 6 cm as indeterminate, and tumors ≥6 cm as having a high risk of malignancy. Multiple international society guidelines continue to use a cut-off size of 4 cm to determine management. For indeterminate nodules 1–4 cm in size on CT, either immediate additional imaging (adrenal protocol CT or magnetic resonance imaging [MRI]) or interval imaging in 6–12 months is recommended due to a slightly increased risk of malignancy. Nodules ≥4 cm with indeterminate imaging characteristics are typically recommended for surgical resection
The Landmark Series: Evaluation and Management of Adrenal Incidentalomas.
Owei L, Wachtel H.
Ann Surg Oncol. 2025 Apr 30. doi: 10.1245/s10434-025-17296-8. Epub ahead of print. PMID: 40304946. - While adrenal incidentalomas are both common and increasing in incidence, there remains a significant gap in understanding and management. Notably, rates of completion of appropriate testing and imaging are low. A systematic review examining adherence to guidelines found that median rates of imaging (34%) and biochemical evaluation (18%) were unacceptably low. In this study, the factor most consistently associated with improved follow-up was the radiologist’s recommendation, highlighting the significance of radiology reporting and suggesting a potential opportunity for improvement.
The Landmark Series: Evaluation and Management of Adrenal Incidentalomas.
Owei L, Wachtel H.
Ann Surg Oncol. 2025 Apr 30. doi: 10.1245/s10434-025-17296-8. Epub ahead of print. PMID: 40304946.
- Results: The CT score helped discriminate between patients with poor prognosis and patients with good prognosis in both the validation cohort (54 patients; mean OS, 69.4 months; 95% confidence interval [CI]: 57.4-81.4 months vs mean OS, 75.6 months; 95% CI: 62.9-88.4 months, respectively; P = .022).In the validation cohort the C-index of the CT score was significantly better than that of the ENSAT stage alone (0.62 vs 0.35; P = .002). Conclusion: A CT score combining morphological criteria, radiomics, and ENSAT stage on preoperative CT examinations allows a better prognostic stratification of patients with ACC compared to ENSAT stage alone.
A Computed Tomography-Based Score to Predict Survival in Patients With Adrenocortical Carcinoma: A Proof-of-Concept Study.
Barat M, Eltaher M, Moawad AW et al.
Can Assoc Radiol J. 2025 May 4:8465371251335170. - Purpose: Adrenocortical carcinoma (ACC) is a rare condition with a poor and hardly predictable prognosis. This study aims to build and evaluate a preoperative computed tomography (CT)-based score (CT score) using features previously reported as biomarkers in ACC to predict overall survival (OS) in patients with ACC. Methods: A CT score based on preoperative CT examinations combining shape elongation, maximum tumour diameter, and the European Network for the Study of Adrenal Tumors (ENSAT) stage was built using a logistic regression model to predict OS duration in a development cohort of 89 patients with ACC. An optimal cut-off of the CT score was defined and the Kaplan-Meier method was used to assess OS. The CT score was then tested in an external validation cohort of 54 patients wit ACC. The C-index of the CT score for predicting OS was compared to that of ENSAT stage alone.
A Computed Tomography-Based Score to Predict Survival in Patients With Adrenocortical Carcinoma: A Proof-of-Concept Study.
Barat M, Eltaher M, Moawad AW et al.
Can Assoc Radiol J. 2025 May 4:8465371251335170. - Adrenocortical carcinoma (ACC) is a rare condition with an estimated incidence of 0.5 to 2 ACCs per million of inhabitants per year, and it accounts for 0.04% to 0.2% of all cancer deaths in the United States. Patients with ACC, generally have a poor prognosis, with a 5-year overall survival (OS) rate of only about 40%. However, OS varies considerably among patients. In patients with ACC, prognostic factors for OS include clinical, histopathological, and molecular features; these factors could help determine appropriate management based on patient prognosis, but histopathological and molecular information are obtained at the penalty of invasive tissue sampling, which is not free of morbidity and carries a risk of tumour dissemination. Therefore, adrenal biopsy is not recommended in the setting of ACC.
A Computed Tomography-Based Score to Predict Survival in Patients With Adrenocortical Carcinoma: A Proof-of-Concept Study.
Barat M, Eltaher M, Moawad AW et al.
Can Assoc Radiol J. 2025 May 4:8465371251335170. - Shape-based radiomic features have demonstrated utility for tumour characterization in oncologic imaging and for prognostic stratification of patients.30 Moreover, a high interobserver reproducibility between radiologists has been reported.30 A shape-based classification using 3 features (roughness, convexity, and sphericity) proved to be able to discriminate between lung granuloma and lung carcinoma with an AUROC of 0.72, which was equivalent to that of expert radiologists. Similarly, Alvarez-Jimenez et al showed that first-order and shape-based radiomic features measured on T2-weighted MRI images of the rectal wall and peritumoral environment of rectal cancer after neoadjuvant radiochemotherapy were effective for the restaging of rectal tumours before surgery.
A Computed Tomography-Based Score to Predict Survival in Patients With Adrenocortical Carcinoma: A Proof-of-Concept Study.
Barat M, Eltaher M, Moawad AW et al.
Can Assoc Radiol J. 2025 May 4:8465371251335170. - ”Our study has some limitations. First, despite a two-centre design, the number of patients remains limited, mostly because of the rarity of ACC, which makes the performance of large studies difficult. However, our study includes a relatively large number of patients compared to previous published studies, as well as an external validation cohort.20 Second, the retrospective design of the study may introduce selection bias, but patients with this rare disease are mostly treated in reference centres that belong to the COMETEcancer network, so their treatment is largely homogeneous. Similarly, the acquisition parameters of the 2 centres were different, which might have influenced the results of the extracted radiomic features; however, the use of voxel size normalization for radiomic analysis should mitigate this risk.”
A Computed Tomography-Based Score to Predict Survival in Patients With Adrenocortical Carcinoma: A Proof-of-Concept Study.
Barat M, Eltaher M, Moawad AW et al.
Can Assoc Radiol J. 2025 May 4:8465371251335170.
Contrast
- Key outcomes from a multidisciplinary task force on hypersensitivity reactions to iodinated contrast media include recommendations to document reactions thoroughly in the electronic health record, including symptoms and the specific inciting agent, and a discussion of varying strategies for avoidance of repeat acute hypersensitivity reactions to iodinated contrast media according to the severity of the index reaction; importantly, no corticosteroid premedication is generally recommended for patients with a prior mild acute hypersensitivity reaction.
Management and Prevention of Hypersensitivity Reactions to Radiocontrast Media: A Consensus Statement From the American College of Radiology and the AAAAI.
Wang C, Ramsey A, Lang D, et al.
J Allergy Clin Immunol Pract. 2025 Mar 10:S2213-2198(25)00191-6. doi: 10.1016/j.jaip.2025.01.042. Epub ahead of print. - For patients with a history of mild immediate ICM hypersensitivity reactions, premedication is not recommended; this is a change from prior American College of Radiology recommendations. Switching the contrast agent is recommended when the inciting agent(s) is known and when feasible.
Hypersensitivity Reactions to Radiocontrast Media: A Consensus Statement From the American College of Radiology and the AAAAI.
Wang C, Ramsey A, Lang D, et al.
J Allergy Clin Immunol Pract. 2025 Mar 10:S2213-2198(25)00191-6. doi: 10.1016/j.jaip.2025.01.042. Epub ahead of print. - For patients with a history of severe immediate ICM hypersensitivity reactions, it is recommended first to consider alternative imaging studies. If there is no acceptable alternative study that does not entail exposure to the same class of contrast, premedication is recommended and switching the contrast agent is recommended when feasible; this is a change from the most recent Joint Task Force Practice Parameters on Anaphylaxis. The study should be performed in a hospital setting with a rapid response team available, including personnel, equipment, and supplies to treat anaphylaxis.
Hypersensitivity Reactions to Radiocontrast Media: A Consensus Statement From the American College of Radiology and the AAAAI.
Wang C, Ramsey A, Lang D, et al.
J Allergy Clin Immunol Pract. 2025 Mar 10:S2213-2198(25)00191-6. doi: 10.1016/j.jaip.2025.01.042. Epub ahead of print. - No premedication is necessary for patients with prior chemotoxic or physiologic reactions or an isolated history of shellfish allergy or iodine allergy including topical povidone-iodine.
Hypersensitivity Reactions to Radiocontrast Media: A Consensus Statement From the American College of Radiology and the AAAAI.
Wang C, Ramsey A, Lang D, et al.
J Allergy Clin Immunol Pract. 2025 Mar 10:S2213-2198(25)00191-6. doi: 10.1016/j.jaip.2025.01.042. Epub ahead of print. - ”The risk of adverse immediate ICM reactions has been dramatically reduced with the universal use of LOCM. There is no high-quality evidence supporting the benefit of corticosteroid premedication in preventing recurrent reactions in patients receiving LOCM, owing to variations in premedication protocols and the low rate of severe reactions to LOCM. Despite these unproven and modest benefits, a survey of radiologists in 2009 showed increasing support for using premedication regimens compared with 1995.”
Hypersensitivity Reactions to Radiocontrast Media: A Consensus Statement From the American College of Radiology and the AAAAI.
Wang C, Ramsey A, Lang D, et al.
J Allergy Clin Immunol Pract. 2025 Mar 10:S2213-2198(25)00191-6. doi: 10.1016/j.jaip.2025.01.042. Epub ahead of print. - “This document contains joint consensus statements endorsed by the ACR and the AAAAI, which are intended to improve and standardize the care of patients who experience or have a history of an adverse reaction to ICM. These consensus recommendations are based on the best evidence and apply only to intravenous administration of ICM. High-quality evidence and methodologically rigorous studies are lacking owing to (1) the rarity of moderate and severe reactions to low-osmolality iodinated contrast agents; (2) the paucity of methodologically sound studies; and (3) the heterogeneity of published studies, including the multiplicity of premedication and ST regimens, variations in patient selection for premedication, and differing contrast agents used in switching methodology. These recommendations should not be taken as definitive standards of practice; they may be subject to change once additional and more definitive evidence becomes available.”
Hypersensitivity Reactions to Radiocontrast Media: A Consensus Statement From the American College of Radiology and the AAAAI.
Wang C, Ramsey A, Lang D, et al.
J Allergy Clin Immunol Pract. 2025 Mar 10:S2213-2198(25)00191-6. doi: 10.1016/j.jaip.2025.01.042. Epub ahead of print. - - Documentation of iodinated contrast media (ICM) hypersensitivity reactions, including symptoms and the specific inciting agent in the electronic medical record, is recommended to optimize future ICM reaction management.
- High-quality evidence and methodologically rigorous studies are lacking owing to: (a) the rarity of moderate and severe reactions to low-osmolality iodinated contrast agents; (b) the paucity of methodologically sound studies; and (c) the heterogeneity of published studies, including the multiplicity of premedication and skin testing regimens, variations in patient selection for premedication, and differing contrast agents used in switching methodology.
Management and Prevention of Hypersensitivity Reactions to Radiocontrast Media: A Consensus Statement From the American College of Radiology and the AAAAI.
Wang C, Ramsey A, Lang D, et al.
J Allergy Clin Immunol Pract. 2025 Mar 10:S2213-2198(25)00191-6. doi: 10.1016/j.jaip.2025.01.042. Epub ahead of print.
Deep Learning
- Results: The CT score helped discriminate between patients with poor prognosis and patients with good prognosis in both the validation cohort (54 patients; mean OS, 69.4 months; 95% confidence interval [CI]: 57.4-81.4 months vs mean OS, 75.6 months; 95% CI: 62.9-88.4 months, respectively; P = .022).In the validation cohort the C-index of the CT score was significantly better than that of the ENSAT stage alone (0.62 vs 0.35; P = .002).
Conclusion: A CT score combining morphological criteria, radiomics, and ENSAT stage on preoperative CT examinations allows a better prognostic stratification of patients with ACC compared to ENSAT stage alone.
A Computed Tomography-Based Score to Predict Survival in Patients With Adrenocortical Carcinoma: A Proof-of-Concept Study.
Barat M, Eltaher M, Moawad AW et al.
Can Assoc Radiol J. 2025 May 4:8465371251335170. - Purpose: Adrenocortical carcinoma (ACC) is a rare condition with a poor and hardly predictable prognosis. This study aims to build and evaluate a preoperative computed tomography (CT)-based score (CT score) using features previously reported as biomarkers in ACC to predict overall survival (OS) in patients with ACC. Methods: A CT score based on preoperative CT examinations combining shape elongation, maximum tumour diameter, and the European Network for the Study of Adrenal Tumors (ENSAT) stage was built using a logistic regression model to predict OS duration in a development cohort of 89 patients with ACC. An optimal cut-off of the CT score was defined and the Kaplan-Meier method was used to assess OS. The CT score was then tested in an external validation cohort of 54 patients wit ACC. The C-index of the CT score for predicting OS was compared to that of ENSAT stage alone.
A Computed Tomography-Based Score to Predict Survival in Patients With Adrenocortical Carcinoma: A Proof-of-Concept Study.
Barat M, Eltaher M, Moawad AW et al.
Can Assoc Radiol J. 2025 May 4:8465371251335170. - Adrenocortical carcinoma (ACC) is a rare condition with an estimated incidence of 0.5 to 2 ACCs per million of inhabitants per year, and it accounts for 0.04% to 0.2% of all cancer deaths in the United States. Patients with ACC, generally have a poor prognosis, with a 5-year overall survival (OS) rate of only about 40%. However, OS varies considerably among patients. In patients with ACC, prognostic factors for OS include clinical, histopathological, and molecular features; these factors could help determine appropriate management based on patient prognosis, but histopathological and molecular information are obtained at the penalty of invasive tissue sampling, which is not free of morbidity and carries a risk of tumour dissemination. Therefore, adrenal biopsy is not recommended in the setting of ACC.
A Computed Tomography-Based Score to Predict Survival in Patients With Adrenocortical Carcinoma: A Proof-of-Concept Study.
Barat M, Eltaher M, Moawad AW et al.
Can Assoc Radiol J. 2025 May 4:8465371251335170. - Shape-based radiomic features have demonstrated utility for tumour characterization in oncologic imaging and for prognostic stratification of patients.30 Moreover, a high interobserver reproducibility between radiologists has been reported.30 A shape-based classification using 3 features (roughness, convexity, and sphericity) proved to be able to discriminate between lung granuloma and lung carcinoma with an AUROC of 0.72, which was equivalent to that of expert radiologists. Similarly, Alvarez-Jimenez et al showed that first-order and shape-based radiomic features measured on T2-weighted MRI images of the rectal wall and peritumoral environment of rectal cancer after neoadjuvant radiochemotherapy were effective for the restaging of rectal tumours before surgery.
A Computed Tomography-Based Score to Predict Survival in Patients With Adrenocortical Carcinoma: A Proof-of-Concept Study.
Barat M, Eltaher M, Moawad AW et al.
Can Assoc Radiol J. 2025 May 4:8465371251335170. - ”Our study has some limitations. First, despite a two-centre design, the number of patients remains limited, mostly because of the rarity of ACC, which makes the performance of large studies difficult. However, our study includes a relatively large number of patients compared to previous published studies, as well as an external validation cohort.20 Second, the retrospective design of the study may introduce selection bias, but patients with this rare disease are mostly treated in reference centres that belong to the COMETEcancer network, so their treatment is largely homogeneous. Similarly, the acquisition parameters of the 2 centres were different, which might have influenced the results of the extracted radiomic features; however, the use of voxel size normalization for radiomic analysis should mitigate this risk.”
A Computed Tomography-Based Score to Predict Survival in Patients With Adrenocortical Carcinoma: A Proof-of-Concept Study.
Barat M, Eltaher M, Moawad AW et al.
Can Assoc Radiol J. 2025 May 4:8465371251335170.
- This analysis found low trust in health care systems to use AI responsibly and protect patients from AI-related harms. General trust in the health care system, but not health literacy or AI knowledge, was associated with these perceptions. This analysis is limited by its observational, cross-sectional design. Future work should examine this trust longitudinally and include additional validated measures of factors such as patient comfort, familiarity, and experience with AI that could be associated with the outcome. Low trust in health care systems to use AI indicates a need for improved communication and investments in organizational trustworthiness.
Patients’ Trust in Health Systems to Use Artificial Intelligence
Paige Nong, PhD; Jodyn Platt, PhD
JAMA Network Open. 2025;8(2):e2460628. doi:10.1001/jamanetworkopen.2024.60628 
Patients’ Trust in Health Systems to Use Artificial Intelligence
Paige Nong, PhD; Jodyn Platt, PhD
JAMA Network Open. 2025;8(2):e2460628. doi:10.1001/jamanetworkopen.2024.60628- The growth and development of artificial intelligence (AI) in health care introduces a new set of questions about patient engagement and whether patients trust systems to use AI responsibly and safely. The answer to this question is embedded in patients’ experiences seeking care and trust in health systems. Meanwhile, the adoption of AI technology outpaces efforts to analyze patient perspectives, which are critical to designing trustworthy AI systems and ensuring patient-centered care. We conducted a national survey of US adults to understand whether they trust their health systems to use AI responsibly and protect them from AI harms. We also examined variables that maybe associated with these attitudes, including knowledge of AI, trust, and experiences of discrimination in health care.
Patients' Trust in Health Systems to Use Artificial Intelligence.
Nong P, Platt J.
JAMA Netw Open. 2025 Feb 3;8(2):e2460628. - This cross-sectional analysis used data from an original survey fielded with a nationally representative sample of US adults via the National Opinion Research Center’s (NORC’s) probability -based AmeriSpeak Panel from June to July 2023. Poststratification survey weights using the Current Population Survey were applied to produce national estimates. Pretesting and cognitive interviews were conducted to ensure content validity and accessibility. Reporting followed the STROBE guideline. NORC obtained participants’ written informed consent, and the study was determined to be unregulated by the University of Michigan IRB.
Patients' Trust in Health Systems to Use Artificial Intelligence.
Nong P, Platt J.
JAMA Netw Open. 2025 Feb 3;8(2):e2460628.
- ”Pancreas segmentation has been traditionally challenging due to its small size in computed tomography abdominal volumes, high variability of shape and positions among patients, and blurred boundaries due to low contrast between the pancreas and surrounding organs. Many deep learning models for pancreas segmentation have been proposed in the past few years. We present a thorough systematic review based on the Preferred Reporting Items for Systematic Reviews and Meta-analyses statement. The literature search was conducted on PubMed, Web of Science, Scopus, and IEEE Xplore on original studies published in peer-reviewed journals from 2013 to 2023. Overall, 130 studies were retrieved. We initially provide an overview of the technical background of the most common network architectures and publicly available datasets.”
Deep learning for pancreas segmentation on computed tomography: a systematic review
Andrea Moglia · Matteo Cavicchioli · Luca Mainard · Pietro Cerveri
Artificial Intelligence Review (2025) 58:220 - “The methods of abdominal organ segmentation can be divided into human intervention-based, target-based, region-based, and learning-based. The first type can be further subdivided into manual, semi-automatic, and automatic methods. Traditionally, medical image segmentation, including pancreas segmentation, has relied heavily on manual delineation by expert radiologists. This poses critical challenges including inter- and intra-observer variability, time-consuming labor, and subjective interpretation. Limited availability of experts, human error, and scalability issues further complicate the process. Extensive training requirements and reproducibility concerns hinder the widespread adoption of manual segmentation methods.”
Deep learning for pancreas segmentation on computed tomography: a systematic review
Andrea Moglia · Matteo Cavicchioli · Luca Mainard · Pietro Cerveri
Artificial Intelligence Review (2025) 58:220 - From a DL perspective, the learning-based models for segmentation face several challenges. First, since supervised learning models need pixel-level annotations they rely on manual annotation from expert clinicians, thus sharing the same challenges as manual annotation. Second, a common problem for semi-supervised learning is the violation of the same statistic distribution for both labeled and unlabeled data. Moreover, the performances can degrade due to incorrectly generated pseudo-labels. Third, for self-supervised learning, designing pretext tasks for medical imaging can be challenging. On the other hand, contrastive learning requires relatively large unlabeled datasets, which are difficult to obtain to maintain patient privacy. Lastly, weakly learning struggles to segment the organ boundaries accurately due to the limited information contained in weak annotations, e.g. scribbles and bounding boxes.
Deep learning for pancreas segmentation on computed tomography: a systematic review
Andrea Moglia · Matteo Cavicchioli · Luca Mainard · Pietro Cerveri
Artificial Intelligence Review (2025) 58:220 - Developers should provide healthcare providers and patients with clear, transparent, and simplified explanations of how DL models work. Patients will also have the right to know how their data will be used, how their privacy will be protected, and the risks of using their data (e.g., AI model bias). More collaboration between developers of DL models and clinicians is encouraged, i.e., by collaborating within projects or in the writing of papers or by attending conferences of mutual interest .
Deep learning for pancreas segmentation on computed tomography: a systematic review
Andrea Moglia · Matteo Cavicchioli · Luca Mainard · Pietro Cerveri
Artificial Intelligence Review (2025) 58:220 - This systematic review of DL applications for segmenting the pancreas and its lesions elucidates significant advancements and identifies important areas of improvement. The review highlights several critical challenges. From a clinical point of view there is an urgent need to improve segmentation of pancreas cancer, given its aggressive nature, and its highly complex surgical treatment. Other clinical applications concern the segmentation of pancreas subregions, especially the head where about two thirds of pancreas cancers occur. From a technical point of view, emphasis should be placed more on method configuration than architectural variations, following the successful paradigm of the nnU-Net architecture. Moreover, the DL models should demonstrate competitive or superior performances over state-of-the-art not just on a single (and small) dataset, but on a wide range of datasets. Model size should also be considered when comparing different approaches to avoid bias in claiming that one outperformed others, which may be smaller.
Deep learning for pancreas segmentation on computed tomography: a systematic review
Andrea Moglia · Matteo Cavicchioli · Luca Mainard · Pietro Cerveri
Artificial Intelligence Review (2025) 58:220
Kidney
- Bosniak Cysts Classification
Class I
Class II
Class IIF
Class III
Class IV - Bosniak Cyst Class I
- well-defined thin (≤2 mm) smooth wall
- homogeneous simple fluid (-9 to 20 HU)
- no septa or calcification
- the wall may enhance after the administration of contrast - Bosniak Cyst Class II
- six types, all with thin (≤2 mm) smooth walls
--- cystic masses with thin (≤2 mm) and few (1-3) septa
----- septa and wall may enhance
----- may have calcifications of any type (although if the calcification is thick and nodular, consider MRI before assigning a class)
--- homogeneous hyperattenuating (≥70 HU) masses on non-contrast CT
--- homogeneous non-enhancing masses >20 HU at renal mass protocol CT
----- may have calcifications of any type (although if the calcification is thick and nodular, consider MRI before assigning a class)
--- homogeneous masses (-9 to 20 HU at non-contrast CT)
--- homogeneous masses (21 to 30 HU at portal venous phase CT)
--- homogeneous low attenuation masses that are too small to characterize - Bosniak Cyst Class IIF
- smooth minimally thickened (3 mm) enhancing wall
- smooth minimal thickening (3 mm) of one or more enhancing septa
- many (≥4) smooth thin (≤2 mm) enhancing septa - Bosniak Cyst Class III
one or more walls or septa that are
- enhancing thick (≥4 mm width)
- enhancing irregular (displaying ≤3 mm obtusely margined convex protrusion[s]) - Bosniak Cyst Class IV
one or more enhancing nodule(s)
- ≥4 mm convex protrusion with obtuse margins
- a convex protrusion of any size that has acute margins - Managing the Bosniak Cyst
- Class I
"Benign simple renal cyst requiring no follow up"
- Class II
"Benign Bosniak II renal cyst requiring no follow up"
"Likely benign Bosniak II renal mass requiring no follow up" - Managing the Bosniak Cyst
- Class IIF
"Bosniak IIF cystic renal mass. The large majority of Bosniak IIF masses are benign. When malignant, nearly all are indolent. Generally, Bosniak IIF masses are followed at imaging at 6 months and 12 months, then annually for a total of 5 years to assess for morphologic change" - Managing the Bosniak Cyst
- Class III
"Bosniak III cystic renal mass. Bosniak III masses have an intermediate probability of being malignant. If not already obtained, consider urology consultation"
- Class IV
"Bosniak IV cystic renal mass. The large majority (up to 90%) of Bosniak IV masses are malignant. If not already obtained, consider urology consultation"
- “The 3 dominant histological subtypes are clear cell (75%- 80%), papillary (10%-15%), and chromophobe (5%).15 Clear cell RCC is named for its golden yellow clear cytoplasm.15 Loss of the von Hippel Lindau (VHL) tumor suppressor gene occurs in up to 90% of clear cell RCC tumors, with VHL inactivation leading to activation of hypoxia and angiogenesis pathways. VHL inactivation results in highly vascular tumors with a high risk of bleeding. VHL inactivation usually occurs through a combination of a deleterious variant and loss of a portion of chromosome 3p, home to VHL, as well as several other common genetic variants.”
Renal Cell Carcinoma: A Review
Tracy L. Rose, William Y. Kim
JAMA. 2024;332(12):1001-1010 - The classic triad of flank pain, a palpable abdominal mass, and hematuria occurs in less than 10% of patients with newly diagnosed RCC.23 Because the retroperitoneal space can accommodate substantial tumor growth prior to symptom onset, only large RCCs are detected by palpation. Currently, the widespread use of abdominal imaging leads to incidental RCC detection in 37% to 61% of cases. With increased incidental detection, gross hematuria is currently reported in less than 25% of patients and occurs more often in advanced disease. Approximately 1.3% of patients with gross hematuria are diagnosed with RCC.
Renal Cell Carcinoma: A Review
Tracy L. Rose, William Y. Kim
JAMA. 2024;332(12):1001-1010 - Paraneoplastic syndromes occur in 10% to 40%of patients with RCC and are not consistently associated with higher stage or grade across studies. Common paraneoplastic manifestations include fever (8%) hypercalcemia (1%-30%), anemia (22%-52%), thrombocytosis (8%-12%), erythrocytosis (2%-4%), and hypertension (3%-18%).Paraneoplastic erythrocytosis is associated with elevated erythropoietin levels, which is produced by RCC tumor cells upon VHL inactivation. Stauffer syndrome, a paraneoplastic syndrome first described in 1961, is characterized by elevated liver enzymes in approximately 3% of patients with RCC; hepatosplenomegaly without liver metastases can also occur with this syndrome. Paraneoplastic syndromes may resolve in up to 52% of patients after nephrectomy or systemic treatment of RCC, and persistence of paraneoplastic symptoms after nephrectomy may indicate residual disease.
Renal Cell Carcinoma: A Review
Tracy L. Rose, William Y. Kim
JAMA. 2024;332(12):1001-1010 - Chromophobe RCC is the third most common malignant histology and tends to be larger in size larger size, is heterogeneous, and shows peak absolute HU of 60 to 80 in the nephrographic and excretory phase, which is in contrast to above three lesions. Although these enhancement patterns are distinct for the majority of lesions, there is overlap in up to 20% of several characteristics among the five different renal masses that pose a challenge to a high-confidence differentiation.
Radiologist’s Disease Imaging for Renal Cancer
Alex Chung, Steven S. Raman
Urol Clin N Am 50 (2023) 161–180 - The most common benign mimic of clear cell RCC is oncocytoma, which has a similar shape and tends to have a more homogeneous enhancement with unenhanced HU between 25 and 40, corticomedullary phase peak absolute HU below 120 and slower washout with nephrographic phase HU and excretory phase enhancement below 100. A subset of both oncocytomas and clear cell RCC have a central scar with delayed enhancement. Another mimic of clear cell RCC is fat-poor AML, which tends to have a lobular, mushroom or ovoid shape lesion often with acute angles to the renal cortex.
Radiologist’s Disease Imaging for Renal Cancer
Alex Chung, Steven S. Raman
Urol Clin N Am 50 (2023) 161–180 - Clear Cell Renal Cell Carcinoma
- Heterogeneous, even at a small size
- Peak enhancement in the corticomedullary phase
- Relative enhancement in corticomedullary phase greater than 0
- Mild T2 hyperintensity
- Can have signal drop out on out-of-phase imaging, due to microscopic fat - Papillary Renal Cell Carcinoma
- More likely to be homogeneous than clear cell RCC
- Peak enhancement in the nephrographic phase
- Relative enhancement less than 0 in all phases
- Restricts diffusion
- T2 hypointense - Chromophobe Renal Cell Carcinoma
- Peak enhancement in the corticomedullary or nephrographic phase
- Relative enhancement less than 0 in all phases - Oncocytoma
- Mimic of clear cell RCC
- Peak enhancement in the corticomedullary phase
- Relative enhancement less than 0 in all phases - Fat-Poor Angiomyolipoma
- Peak enhancement in the corticomedullary phase
- Relative enhancement less than 0 in all phases
- Tends to be > 45 HU on unenhanced CT
- T2 hypointense - There is a clear benefit of imaging-based differentiation of small indeterminate masses to its subtypes of clear cell RCC, chromophobe RCC, papillary RCC, fat poor AML, and oncocytoma because it helps determine the next step options for the patients. The work thus far in radiology has explored different parameters in CT, MRI, and CEUS with the discovery of many reliable imaging features that suggest certain tissue subtypes. Likert score-based risk stratification systems can help determine management, and new techniques such as perfusion, radiogenomics, SPECT, and artificial intelligence can add to imaging-based evaluation of indeterminate renal masses.
Radiologist’s Disease Imaging for Renal Cancer
Alex Chung, Steven S. Raman
Urol Clin N Am 50 (2023) 161–180 - “Chromophobe renal cell carcinoma (ChRCC) is the second most common form of non-clear cell renal cell carcinoma after papillary RCC and accounts for 5-10% of all kidney cancers. The pathogenesis of ChRCC suggests that this subtype is derived from cells of the distal convoluted tubules of the nephron, in contrast to clear cell renal cell carcinoma, which arises from the proximal tubules. Despite the presentation of ChRCC as large tumors, studies from various surgical cohorts have shown more favorable clinical courses than for the other RCC subtypes, suggesting a low metastatic potential.”
Chromophobe Renal Cell Carcinoma: Results From a Large Single-Institution Series.
Casuscelli J, Becerra MF, Seier K,
Clin Genitourin Cancer. 2019 Oct;17(5):373-379. - In the present study we confirmed that chromophobe renal cell carcinoma is associated with a more favorable clinical outcome and a lower propensity for metastatic development than clear cell RCC, but we were also able to identify risk factors associated with metastatic development. Increased size and sarcomatoid differentiation of ChRCC tumors lead to lower RFS and OS. Interestingly, in our large series chromophobe renal cell carcinoma was diagnosed more often in females and younger patients compared with clear cell RCC. We further observed that in clear cell RCC, lower BMI and male gender were associated with decreased survival, but this was not the case for chromophobe renal cell carcinoma.
Chromophobe Renal Cell Carcinoma: Results From a Large Single-Institution Series.
Casuscelli J, Becerra MF, Seier K,
Clin Genitourin Cancer. 2019 Oct;17(5):373-379. - Chromophobe renal cell carcinoma (ChRCC) is the second most common variant histology (non-clear cell) RCC. ChRCC is distinct from clear cell RCC (ccRCC) in terms of genetics, genomics, metabolism, cell of origin, and response to targeted and immune therapies. The pathogenesis of ChRCC remains unclear, but current data suggest two potential mechanisms: mTORC1 hyperactivation through PTEN pathway mutations and mitochondrial dysfunction leading to oxidative stress. There are no specific approved treatments for ChRCC, although some responses to tyrosine kinase and mTOR inhibitors have been observed. Response to immunotherapy is generally limited. Targetable pathways involving innate lymphoid cells/IL-15 and cysteine homeostasis/ferroptosis have recently been identified.
Chromophobe renal cell carcinoma
Elizabeth P. Henske et al.
Cancer Cell Volume 41, Issue 8p1383-1388August 14, 2023 
- Renal cell carcinoma (RCC) is a heterogeneous group of neoplasms derived from the renal tubular epithelial cells. Chromophobe RCC(chRCC) is the third most common subtype of RCC, accounting for 5% of cases. chRCC may be detected as an incidental finding or less commonly, may manifest with clinical symptoms. The mainstay of therapy for chRCC is surgical resection. ChRCC has a better prognosis compared with the more common clear cell RCC. At gross pathologic analysis, chRCC is a solid, well-defined mass with lobulated borders.
Chromophobe Renal Cell Carcinoma with Radiologic-Pathologic Correlation orders.
Jamie Marko, Ryan Craig, Andrew Nguyen,et al. RadioGraphics 2021; 41:1408–1419 - .The most common imaging pattern is a predominantly solid renal mass with circumscribed margins and enhancement less than that of the renal cortex.
Chromophobe Renal Cell Carcinoma with Radiologic-Pathologic Correlation.
Jamie Marko, et al.
RadioGraphics 2021; 41:1408–1419 - “chRCC is often detected as an incidental finding or less commonly manifests with clinical symptoms . With increased use of cross-sectional abdominal imaging, there is a trend toward detection of asymptomatic renal masses, including chRCC. The reported clinical findings of chRCC include local symptoms such as abdominal or flank pain (34%–67% of cases), abdominal mass (27% of cases), and hematuria (17%–40% of cases) and systemic symptoms such as fever, cachexia, fatigue, and weight loss (20%–33% of cases). Most chRCCs are low-stage cancers at diagnosis, reflecting both the indolent nature of chRCC and the trend toward incidental detection at imaging. Lymph node and distant metastases are infrequent, seen in 2%–4% and 1%–4% of cases, respectively. The most common sites of metastases are the liver and lungs.”
Chromophobe Renal Cell Carcinoma with Radiologic-Pathologic Correlation.
Jamie Marko, et al.
RadioGraphics 2021; 41:1408–1419 - “The estimated worldwide incidence of RCC in 2018 was 400000, with 175000 related deaths . Of these 400000 cases worldwide, approximately 65000 occurred in the United States. Since chRCC accounts for approximately 5% of RCC cases, the estimated U.S. annual incidence is 3250 cases. The reported mean age of patients diagnosed with chRCC is 58–60 years. chRCC shows a slight male sex predilection.”
Chromophobe Renal Cell Carcinoma with Radiologic-Pathologic Correlation.
Jamie Marko, et al.
RadioGraphics 2021; 41:1408–1419 - “RCC is seven times more common in patients with BHD than in the general population. Approximately 25%–35% of patients with BHD develop renal tumors, with a mean age at diagnosis of 50 years. The RCCs that occur in BHD are frequently bilateral, multifocal, and slow growing. Approximately 90% of the renal tumors seen in BHD are either hybrid oncocytic tumors (mixed oncocytoma and chRCC) or pure chRCCs. Other renal tumors seen in patients with BHD include oncocytoma, ccRCC, and papRCC.”
Chromophobe Renal Cell Carcinoma with Radiologic-Pathologic Correlation.
Jamie Marko, et al.
RadioGraphics 2021; 41:1408–1419 - “chRCCs vary in size, with mean sizes ranging from 3.2 cm to 7.0 cm reported in radiologic case series. They are usually confined to the kidney, with local invasion or renal vein invasion uncommonly identified at imaging. Nearly all lesions are solid or mostly solid and sharply marginated. They may be homogeneous or heterogeneous, with the heterogeneous appearance predominating. Calcification and/or a central scar with or without a spoke-wheel pattern of enhancement is seen in a significant minority of cases. A dominant cystic component, hemorrhage, fat, and microscopic lipid are very uncommon findings. chRCC is a moderately vascular tumor. It enhances less than the renal cortex in all phases, as expected given the lack of prominent vascularity seen pathologically. In addition, peak enhancement occurs in the nephrographic phase.”
Chromophobe Renal Cell Carcinoma with Radiologic-Pathologic Correlation.
Jamie Marko, et al.
RadioGraphics 2021; 41:1408–1419 - “At CT, chRCC is typically a well-circumscribed mass with smooth or lobular contours. Most chRCCs are solid or mostly solid at CT. chRCCs may be homogeneous or heterogeneous, with the heterogeneous appearance more commonly reported. Calcification is seen in 14%–34% of cases. A central scar is present in 19%–34% of cases, with a spoke-wheel pattern of enhancement described in a minority of cases . On noncontrast CT images, chRCC tends to be isoattenuating to slightly hyperattenuating compared with the background renal parenchyma . chRCC shows moderate enhancement that peaks in the nephrographic phase or less commonly in the corticomedullary phase. Enhancement is less than that of the renal cortex during all phases. chRCC enhances less than ccRCC but more than papRCC.”
Chromophobe Renal Cell Carcinoma with Radiologic-Pathologic Correlation.
Jamie Marko, et al.
RadioGraphics 2021; 41:1408–1419 - “No single imaging finding is pathognomonic for chRCC; however, understanding the key imaging features of this cancer allows the radiologist to reliably distinguish chRCC from many important renal lesions that are encountered regularly in clinical practice . First, chRCCs are solid or nearly entirely solid lesions without significant regions of necrosis, hemorrhage, or cystic change . This finding allows easy differentiation of chRCC from both low– and high–Bosniak-class cystic lesions, including simple cysts, hemorrhagic cysts, multilocular cystic renal neoplasm of low malignant potential, and the mixed epithelial and stromal family of tumors.”
Chromophobe Renal Cell Carcinoma with Radiologic-Pathologic Correlation.
Jamie Marko, et al.
RadioGraphics 2021; 41:1408–1419 - The differentiation of chRCC from the more common ccRCC relies largely on the reproducible differences in enhancement. At both CT and MRI, chRCC tends to enhance less than the renal cortex in all phases and shows peak enhancement in the nephrographic phase. ccRCC enhances earlier and more intensely, with peak enhancement greater than that of the cortex in the corticomedullary phase. In addition, chRCC is usually localized to the kidney and lacks necrosis, hemorrhage, or cystic change, while ccRCC more commonly shows aggressive behavior such as perinephric fat and/or renal vein invasion, distant metastases, or tumor necrosis .
Chromophobe Renal Cell Carcinoma with Radiologic-Pathologic Correlation.
Jamie Marko, et al.
RadioGraphics 2021; 41:1408–1419 - “The clinical outcomes of patients with chRCC are good, with higher rates of recurrence-free survival (RFS), cancer-specific survival, and overall survival compared with those for patients with ccRCC The rates of RFS in chRCC are 89% and 79% at 5 and 10 years, respectively. Despite the good prognosis associated with most chRCCs, large tumor size, high T stage, small or large vessel vascular invasion, sarcomatoid differentiation, and tumor necrosis have been identified as negative prognostic indicators in a minority of cases. Unlike ccRCCs and papRCCs, chRCCs are not routinely graded owing to their inherent nuclear atypia.”
Chromophobe Renal Cell Carcinoma with Radiologic-Pathologic Correlation.
Jamie Marko, et al.
RadioGraphics 2021; 41:1408–1419 - “chRCC is the third most common subtype of RCC. Although it may occur in association with BHD, it frequently develops sporadically and is often an incidental finding discovered at imaging. The pathologic features, including well-circumscribed margins, moderate vascularity, and lack of internal fat or a significant cystic component, predict the imaging findings of a solid well-circumscribed mass with enhancement less than that of the renal cortex in all phases. Knowledge of the clinically indolent behavior and expected imaging pattern of chRCC will aid the radiologist in the evaluation of a solid renal mass detected at imaging.”
Chromophobe Renal Cell Carcinoma with Radiologic-Pathologic Correlation.
Jamie Marko, et al.
RadioGraphics 2021; 41:1408–1419 
- Sixty percent of the patients were men, with a mean age of 60.2 years. Forty-six percent of cases were incidentally identified, without patient symptoms. None of the patients had evidence of distant metastatic disease, either on initial staging CT or over the course of follow-up (mean, 2.0 years). Mean maximal tumor diameter was 5.24 cm. Forty-six percent of tumors were homogeneous, 85% of lesions were either completely solid or mostly solid, 14% showed calcifications, and 34% showed a central scar or necrosis. Mean maximum attenuation values were 87.9 HU (arterial phase), 83.9 HU (venous phase), and 60.6 HU (delayed phase), with an average delayed washout of 31%. Tumor-to-cortex ratios for the three enhanced phases were 0.59, 0.48, and 0.50, respectively.
Chromophobe renal cell carcinoma: multiphase MDCT enhancement patterns and morphologic features.
Raman SP, Johnson PT, Allaf ME, Netto G, Fishman EK..
AJR 2013 Dec;201(6):1268-76 - Chromophobe RCCs were found to have a wider variability of CT features than previously reported, although they do have a greater propensity for homogeneity and the presence of a central scar or necrosis. Their enhancement characteristics fall in between those of clear cell and papillary RCC, although there is considerable overlap.
Chromophobe renal cell carcinoma: multiphase MDCT enhancement patterns and morphologic features.
Raman SP, Johnson PT, Allaf ME, Netto G, Fishman EK..
AJR 2013 Dec;201(6):1268-76 - From a morphologic perspective on CT, although some series have suggested that chromophobe RCCs may have a higher rate of calcification (as high as 38% according to Kim et al., only 14% of chromophobe RCCs in our series had calcification, comparable to rates previously noted for clear cell RCCs. The vast majority of lesions were either completely solid or mostly solid (85%), whereas only 3% were predominantly cystic (with small solid enhancing components) and 12% were equally cystic and solid. This is broadly concordant with rates reported for RCCs overall (all subtypes), which are thought to present as primary cystic lesions in 4–15% of cases.
Chromophobe renal cell carcinoma: multiphase MDCT enhancement patterns and morphologic features.
Raman SP, Johnson PT, Allaf ME, Netto G, Fishman EK..
AJR 2013 Dec;201(6):1268-76 - In general, the chromophobe RCCs in our series appeared to be hypoenhancing compared with clear cell RCCs in other series, with maximum attenuation values of 87.9 HU (54.6 HU enhancement), 83.9 HU (51.4 HU enhancement), and 60.6 HU (27.3 HU enhancement) in the arterial, venous, and delayed phases, respectively. These values were comparable to those from a study by Young et al., who reported attenuation values of 78.5, 89.5, and 63.0 HU in the corticomedullary, nephrographic, and excretory phases, respectively. Unlike their series, however, the attenuation values for the tumors in our study peaked in the arterial phase, rather than the nephrographic phase.
Chromophobe renal cell carcinoma: multiphase MDCT enhancement patterns and morphologic features.
Raman SP, Johnson PT, Allaf ME, Netto G, Fishman EK..
AJR 2013 Dec;201(6):1268-76 - Chromophobe RCCs are most likely to present as a well-circumscribed relatively homogeneous mass, often with a central scar and an enhancement pattern that is hypovascular relative to clear cell RCC and mildly hypervascular relative to papillary variants. In some cases, raising the possibility of this diagnosis prospectively may be clinically useful, because nephron-sparing surgery should certainly be considered for these tumors given their almost universally positive prognosis. Although none of the tumors in our series had evidence of distant metastatic disease at presentation or during the average 2-year follow-up period, longer-term follow-up studies are warranted to determine whether features such as necrosis and cystic components on IV contrast-enhanced CT are associated with disease recurrence, given the propensity of RCC for late presentation of metastatic disease or recurrence.
Chromophobe renal cell carcinoma: multiphase MDCT enhancement patterns and morphologic features.
Raman SP, Johnson PT, Allaf ME, Netto G, Fishman EK..
AJR 2013 Dec;201(6):1268-76 - ” Chromophobe RCCs reportedly have the best prognosis of all of the different RCC subtypes, with a 5-year survival rate of over 90%, as opposed to clear cell and papillary RCCs, which have rates of survival of 55– 60% and 80–90%, respectively. This subtype of RCC is unusual histopathologically, as evidenced by the fact that Fuhrman grading (which is commonly used to histologically grade “conventional” RCCs) is not used in the histopathologic evaluation of chromophobe RCCs and is not thought to be predictive of ultimate clinical outcomes.”
Chromophobe renal cell carcinoma: multiphase MDCT enhancement patterns and morphologic features.
Raman SP, Johnson PT, Allaf ME, Netto G, Fishman EK.
AJR Am J Roentgenol. 2013 Dec;201(6):1268-76. - ” Although there is a greater variability in the morphologic appearance and enhancement characteristics of chromophobe RCCs than previously described in the literature, with substantial overlap in the tumor’s appearance compared with clear cell and papillary variants, certain features in a lesion’s appearance should at least raise the possibility of a chromophobe RCC. Our retrospective study of 35 patients imaged with 16-, 64-, or 128-MDCT suggests that chromophobe RCCs are most likely to present as a well-circumscribed relatively homogeneous mass, often with a central scar and an enhancement pattern that is hypovascular relative to clear cell RCC and mildly hypervascular relative to papillary variants. In some cases, raising the possibility of this diagnosis prospectively may be clinically useful, because nephron-sparing surgery should certainly be considered for these tumors given their almost universally positive prognosis.”
Chromophobe renal cell carcinoma: multiphase MDCT enhancement patterns and morphologic features.
Raman SP, Johnson PT, Allaf ME, Netto G, Fishman EK.
AJR Am J Roentgenol. 2013 Dec;201(6):1268-76.
Musculoskeletal
- Imaging of chest wall disorders
- congenital and developmental anomalies
- inflammatory and infectious diseases
- soft-tissue and bone tumors
Pancreas
- ”Pancreas segmentation has been traditionally challenging due to its small size in computed tomography abdominal volumes, high variability of shape and positions among patients, and blurred boundaries due to low contrast between the pancreas and surrounding organs. Many deep learning models for pancreas segmentation have been proposed in the past few years. We present a thorough systematic review based on the Preferred Reporting Items for Systematic Reviews and Meta-analyses statement. The literature search was conducted on PubMed, Web of Science, Scopus, and IEEE Xplore on original studies published in peer-reviewed journals from 2013 to 2023. Overall, 130 studies were retrieved. We initially provide an overview of the technical background of the most common network architectures and publicly available datasets.”
Deep learning for pancreas segmentation on computed tomography: a systematic review
Andrea Moglia · Matteo Cavicchioli · Luca Mainard · Pietro Cerveri
Artificial Intelligence Review (2025) 58:220 - “The methods of abdominal organ segmentation can be divided into human intervention-based, target-based, region-based, and learning-based. The first type can be further subdivided into manual, semi-automatic, and automatic methods. Traditionally, medical image segmentation, including pancreas segmentation, has relied heavily on manual delineation by expert radiologists. This poses critical challenges including inter- and intra-observer variability, time-consuming labor, and subjective interpretation. Limited availability of experts, human error, and scalability issues further complicate the process. Extensive training requirements and reproducibility concerns hinder the widespread adoption of manual segmentation methods.”
Deep learning for pancreas segmentation on computed tomography: a systematic review
Andrea Moglia · Matteo Cavicchioli · Luca Mainard · Pietro Cerveri
Artificial Intelligence Review (2025) 58:220 - From a DL perspective, the learning-based models for segmentation face several challenges. First, since supervised learning models need pixel-level annotations they rely on manual annotation from expert clinicians, thus sharing the same challenges as manual annotation. Second, a common problem for semi-supervised learning is the violation of the same statistic distribution for both labeled and unlabeled data. Moreover, the performances can degrade due to incorrectly generated pseudo-labels. Third, for self-supervised learning, designing pretext tasks for medical imaging can be challenging. On the other hand, contrastive learning requires relatively large unlabeled datasets, which are difficult to obtain to maintain patient privacy. Lastly, weakly learning struggles to segment the organ boundaries accurately due to the limited information contained in weak annotations, e.g. scribbles and bounding boxes.
Deep learning for pancreas segmentation on computed tomography: a systematic review
Andrea Moglia · Matteo Cavicchioli · Luca Mainard · Pietro Cerveri
Artificial Intelligence Review (2025) 58:220 - Developers should provide healthcare providers and patients with clear, transparent, and simplified explanations of how DL models work. Patients will also have the right to know how their data will be used, how their privacy will be protected, and the risks of using their data (e.g., AI model bias). More collaboration between developers of DL models and clinicians is encouraged, i.e., by collaborating within projects or in the writing of papers or by attending conferences of mutual interest .
Deep learning for pancreas segmentation on computed tomography: a systematic review
Andrea Moglia · Matteo Cavicchioli · Luca Mainard · Pietro Cerveri
Artificial Intelligence Review (2025) 58:220 - This systematic review of DL applications for segmenting the pancreas and its lesions elucidates significant advancements and identifies important areas of improvement. The review highlights several critical challenges. From a clinical point of view there is an urgent need to improve segmentation of pancreas cancer, given its aggressive nature, and its highly complex surgical treatment. Other clinical applications concern the segmentation of pancreas subregions, especially the head where about two thirds of pancreas cancers occur. From a technical point of view, emphasis should be placed more on method configuration than architectural variations, following the successful paradigm of the nnU-Net architecture. Moreover, the DL models should demonstrate competitive or superior performances over state-of-the-art not just on a single (and small) dataset, but on a wide range of datasets. Model size should also be considered when comparing different approaches to avoid bias in claiming that one outperformed others, which may be smaller.
Deep learning for pancreas segmentation on computed tomography: a systematic review
Andrea Moglia · Matteo Cavicchioli · Luca Mainard · Pietro Cerveri
Artificial Intelligence Review (2025) 58:220
- In this study, stable cysts during the initial 5-year surveillance had a very low risk of malignant conversion compared With those that changed. Surveillance of BD-IPMN with cysts smaller than 20 mm, no WFs or HRS, and no morphologic Changes during the initial 5-year surveillance may be discontinued in elderly patients, those with a limited life expectancy (≤10 years), or those unfit for surgery.
Optimal Surveillance Interval of Branch Duct Intraductal Papillary Mucinous Neoplasm of the Pancreas
Youngmin Han, Wooil Kwon, Mirang Lee, et al.
JAMA Surg. 2024;159(4):389-396 - “These findings suggest that BD-IPMN surveillance may depend on the size of the cyst and morphologic changes at the initial 6-month follow-up. For patients with small cysts (ie, <20 mm) with no morphologic changes during the initial 5-year surveillance period, surveillance may be discontinued for those unfit for surgery or who have a limited life expectancy of 10 years or less.”
Optimal Surveillance Interval of Branch Duct Intraductal Papillary Mucinous Neoplasm of the Pancreas
Youngmin Han, Wooil Kwon, Mirang Lee, et al.
JAMA Surg. 2024;159(4):389-396 - Although most cases of BD-IPMN are indolent and dormant, Some cysts develop malignant features. The findings from this A cohort study suggests that the surveillance protocol should be customized based on morphologic changes and malignant conversion According to the cyst size. We recommend 1.5-, 1-, and 0.5- Yearly surveillance intervals for surveillance of BD-IPMN cysts smaller than 20 mm, 20 to 30 mm, and 30 mm or larger, respectively, after an initial 6-month follow-up. Furthermore, Surveillance may be discontinued for patients with a life expectancy of 10 years or less and who have cysts smaller than 20mm that have remained stable during 5-year surveillance.
Optimal Surveillance Interval of Branch Duct Intraductal Papillary Mucinous Neoplasm of the Pancreas
Youngmin Han, Wooil Kwon, Mirang Lee, et al.
JAMA Surg. 2024;159(4):389-396 - Overview: This study evaluates optimal surveillance intervals for branch duct intraductal papillary mucinous neoplasms (BD-IPMN) of the pancreas.
Importance: Increasing prevalence of intraductal papillary mucinous neoplasm (IPMN) necessitates effective surveillance strategies. Limited data on long-term growth and malignant conversion rates of BD-IPMN.
Objective: Determine optimal surveillance protocols for BD-IPMN. Identify patients who may be exempt from surveillance. Study Design Analyzed data from 3,656 patients with BD-IPMN across five tertiary centers from 1988 to 2020. Follow-up duration averaged 84 months.
Results: 54% of patients were female, average age 63.7 years. 4.7% of patients had malignant lesions confirmed through surgery. Optimal surveillance intervals suggested: 1.5 years for cysts <20 mm, 1 year for 20-30 mm, and 6 months for ≥30 mm.
Conclusions: Surveillance should be tailored based on cyst size and morphological changes. Patients with stable cysts <20 mm may discontinue surveillance if no worrisome features are present after 5 years.
- Pancreatic cancer is frequently a lethal disease with an aggressive tumour biology often presenting with non-specific symptoms. Median survival is approximately 4 months with a 5-year survival of 13%. Surveillance is recommended in individuals with familial pancreatic cancer, specific mutations, and high-risk intraductal papillary mucinous neoplasm, as they are at high risk of developing pancreatic cancer. Chemotherapy combined with surgical resection remains the cornerstone of treatment. However, only a small subset of patients are candidates for surgery. Multi-agent chemotherapy has improved survival in the palliative setting for patients with metastatic disease, as (neo) adjuvant and induction therapy have in patients with borderline resectable and locally advanced pancreatic. Given that pancreatic cancer is predicted to become the second leading cause of cancer-related death by 2030, novel therapies are urgently needed.
Pancreatic cancer
Thomas F Stoop, Ammar A Javed, Atsushi Oba, et al.
Lancet 2025; 405: 1182–202 - Approximately 95% of these tumours arise from precancerous lesions called pancreatic intraepithelial neoplasias. Other precursor lesions are pancreatic cysts, including intraductal papillary mucinous neoplasms (IPMNs) and mucinous cystic neoplasms. In about half of patients with pancreatic cancer and a concomitant IPMN, the invasive tumour is not derived from IPMN. In the general population, pancreatic cysts are estimated to occur in 16% (95% CI 13–18) of individuals, and IPMN is the most common and has the highest risk of malignant transformation. IPMNs with low-risk features has a cumulative incidence of malignant transformation of 8% (4–12) at 10 years, although the risk can be up to 25% (15–36) for patients with high-risk features.
Pancreatic cancer
Thomas F Stoop, Ammar A Javed, Atsushi Oba, et al.
Lancet 2025; 405: 1182–202 - Multiple molecular alterations are observed in pancreatic cancer, which involve activation of oncogenes, inactivation of tumour suppressor genes, and alterations in DNA damage repair genes and homologous repair deficiency genes. Tumorigenesis is entirely somatic in 91% of patients, whereas germline mutations are found in 9% of individuals. The most common somatic mutations are KRAS (~88%), TP53 (61–74%), CDKN2A (16–44%), and SMAD4 (20–22%). The major genetic event in the development of pancreatic ductal adenocarcinoma is the somatic KRAS oncogene mutation, which is observed in more than 90% of patients with low-grade pancreatic intraepithelial neoplasia.
Pancreatic cancer
Thomas F Stoop, Ammar A Javed, Atsushi Oba, et al.
Lancet 2025; 405: 1182–202 - “Diabetes is considered to be another risk factor for pancreatic cancer (RR 1·5, 95% CI 1·4–1·6). However, diabetes can also be a prodrome of pancreatic cancer (ie, type 3c or pancreatogenic diabetes). The fact that diabetes can be both a risk factor and a prodome of pancreatic cancer is illustrated by a possibly stronger association of new-onset diabetes (ie, ≤3–4 years) with pancreatic cancer compared with long-standing diabetes (ie, >3–4 years). Elevated HbA1c in individuals with new-onset diabetes is associated with an increased risk of pancreatic cancer.”
Pancreatic cancer
Thomas F Stoop, Ammar A Javed, Atsushi Oba, et al.
Lancet 2025; 405: 1182–202 - “Pancreatitis is established as a strong risk factor for pancreatic cancer, including chronic pancreatitis (standardised incidence ratio [SIR] 22·6, 95% CI 14·4–35·4), with an increasing cumulative incidence during its disease course. Contrastingly, in patients with acute pancreatitis, the risk of pancreatic cancer is the highest within the first 2 months from symptom onset, as the condition might be a first symptom of pancreatic cancer (HR 172·8, 95% CI 54·9–544·7). The risk gradually decreases over time until the risk disappears 10 years after diagnosis.”
Pancreatic cancer
Thomas F Stoop, Ammar A Javed, Atsushi Oba, et al.
Lancet 2025; 405: 1182–202 - The imaging modality of choice in individuals with suspicion of pancreatic cancer is a contrast-enhanced three-phase (ie, pancreatic, arterial, and portal venous phase) computed tomography (CT), which has a diagnostic accuracy of 89% (95% CI 85–93). Pancreatic cancer in the head often causes dilatation of both the common bile duct and main pancreatic duct (ie, double duct sign). Differentiating between pancreatic cancer and other periampullary carcinomas can be challenging, as illustrated by the 13–32% preoperative misdiagnosis risk among resected pancreatic head adenocarcinomas. Pancreatic adenocarcinoma typically appears as a hypodense lesion on CT, in contrast with hyperdense pancreatic neuroendocrine tumours. However, about 5–17% of lesions are isoattenuating on CT, particularly smaller lesions.
Pancreatic cancer
Thomas F Stoop, Ammar A Javed, Atsushi Oba, et al.
Lancet 2025; 405: 1182–202 - Genetic testing for inherited mutations is recommended by the 2024 NCCN guideline for all patients diagnosed with pancreatic cancer. When tumour-directed treatment is considered in patients with locally advanced and metastatic disease, molecular profiling is recommended to investigate the presence of actionable somatic mutations with therapeutic consequences. These include entities such as BRCA1, BRCA2, DNA mismatch repair deficiency (eg, MLH1, MSH2, MSH6), and KRAS wild type (eg, fusion genes such as NRG and NTRK), despite their low incidences.
Pancreatic cancer
Thomas F Stoop, Ammar A Javed, Atsushi Oba, et al.
Lancet 2025; 405: 1182–202 
Pancreatic cancer
Thomas F Stoop, Ammar A Javed, Atsushi Oba, et al.
Lancet 2025; 405: 1182–202- Approximately 57% of patients with pancreatic cancer present with metastatic disease at the time of diagnosis, with cancer predominantly located in the liver (75–80%), peritoneum (13–30%), lung (15–18%), and extra-regional lymph nodes (12%). In the absence of metastases, the primary tumour is anatomically staged as resectable (RPC), borderline resectable (BRPC), or locally advanced (LAPC), depending on the presence and extent of involvement of peripancreatic major vasculature including the superior mesenteric artery, coeliac axis, hepatic artery branches, and portomesenteric venous axis.
Pancreatic cancer
Thomas F Stoop, Ammar A Javed, Atsushi Oba, et al.
Lancet 2025; 405: 1182–202 - The introduction of multi-agent chemotherapy, either as palliative chemotherapy for metastatic disease or as (neo)adjuvant or induction therapy in patients with localised pancreatic cancer, has improved the survival of patients with pancreatic cancer. Given the prospect that pancreatic cancer will become the second leading cause of cancer-related deaths by 2030, there is an urgent need for novel tumour-targeted therapies with promising results from new generation immunotherapies and KRAS-directed therapies.
Pancreatic cancer
Thomas F Stoop, Ammar A Javed, Atsushi Oba, et al.
Lancet 2025; 405: 1182–202
