Deep learning model enhances 3D segmentation of pancreatic cancer

A newly developed deep learning auto-segmentation model effectively segments pancreatic cancer and surrounding structures to assist surgical planning. In a study involving 275 patients, the model achieved a Dice similarity coefficient of 75.4 in internal and 75.6 in external validation. High segmentation accuracy was noted in the pancreas parenchyma but lower in pancreatic cancer lesions. Notably, cancer detection improved with larger tumor sizes, indicating potential for significant clinical applications in preoperative assessment.