Automatic pancreatic cancer segmentation and classification in CT images using an integrated deep-learning approach
Koteswaramma Dodda, G. MuneeswariAbstract
Background:Successful treatment for pancreatic cancer depends on timely and precise diagnosis because the disease has a low chance of survival. The critical challenge of effectively distinguishing between tumorous and non-tumorous pancreatic tissues in computed tomography scans is pancreatic cancer classification. Using detailed cross-sectional images provided by computed tomography scans, radiologists and oncologists can examine the properties and morphology of the pancreas. Furthermore, deep learning algorithms can obtain precise image analysis and in-depth diagnostic knowledge for therapeutic use.
Methods:This research proposes an integrated artificial intelligence system based on deep learning to segment and classify pancreatic cancer. The tumor-affected region on computed tomography scans can be identified using an Enhanced UNet model segmentation technique. The Modified ResNext model is used to classify pancreatic cancer. Ultimately, the modified ResNext model�s hyper-parameter tuning is achieved using the tunicate swarm optimization algorithm, which helps to increase classification performance. The proposed deep learning models aim to create a reliable and accurate approach to enhance pancreatic cancer diagnosis performance. A benchmark computed tomography image database was used to test the suggested method�s experimental results.
Results:The experimental results show that the proposed Modified ResNext model effectively classifies the pancreatic cancer images into benign and malignant stages with a maximum accuracy of 99.85%, sensitivity of 99.76%, specificity of 99.72%, precision of 99.54%, F-measure of 99.23%, it offers huge possibilities and safety in the automated diagnosing of benign and malignant malignancies. The proposed Enhanced UNet model correctly segments the accurate region of the pancreatic tumor with a higher Intersection Over Union of 96.04% and Dice Similarity Coefficient (DSC) of 95.87%. A comprehensive analysis of the results showed that the proposed strategy performed favorably compared to more cutting-edge techniques. The pancreatic cancer classification and tumor segmentation performance using the proposed integrated model was excellent, indicating its ability to detect pancreatic cancer effectively and precisely.