Deep Learning Algorithms for Pancreas Segmentation from Radiology Scans: A Review

Accurate pancreas segmentation from radiology scans is a critical tool for diagnosing, treating, and monitoring various pancreatic diseases such as pancreatic cancer , diabetes, pancreatitis , and others. Volumetry of the pancreas and precise organ boundary information can help doctors provide better patient care. In recent years, deep learning-based automatic segmentation methods have been proposed to overcome the manual annotation process, which is expensive and tedious. This review paper comprehensively analyzes the current state-of-the-art segmentation strategies in the deep learning era, encompassing techniques spanning from convolutional neural networks (CNNs) to Transformers and their novel and most recent technical configurations as applied to CT and MRI imaging modalities. We also identify the key challenges in the critical sting methodologies and outline potential future directions for advancing pancreas segmentation from radiology scans. The review is presented to serve both the clinical and technical audiences, with each method's key advantages and disadvantages highlighted. Publicly available data sets and free software code are enlisted for reproducible research. We hope our analysis can inspire novel ideas, promote larger-scale collaborations, and eventually lead to significant advances in cutting-edge pancreas segmentation models for clinical applications where diagnosis, treatment planning, and disease monitoring will be streamlined.