Deep-learning based detection model of small pancreatic cystic lesions on contrast-enhanced CT.

With the advancement of radiological technology, the detection rate of pancreatic cystic lesions (PCLs) has been increasingly augmented, particularly the small cysts (<1cm) are being observed more frequently during examinations. In clinical practice, small cysts are prone to being overlooked, which poses a significant challenge for radiologists. Consequently, we propose a deep learning model capable of automatically detecting cysts and segmenting both the pancreas and cysts. The model incorporates dual-phase imaging information and includes a dataset consisting of pancreatic images with diverse characteristics, demonstrating proficient and robust performance in both detection and segmentation.