U-Net-based Pancreas Tumor Segmentation from Abdominal CT Images

There is no doubt that pancreatic cancer is one of the most deadly types of cancer. In order to diagnose and stage pancreatic tumors, computed tomography (CT) is widely used. However, manual segmentation of volumetric CT scans is a time-consuming and subjective process. It has been shown that the U-Net model is highly effective for semantic segmentation, although several deep learning models have been proposed. In this study, we propose a U-Net-based method for pancreatic tumor segmentation from abdominal CT images and demonstrate its simplicity and effectiveness. Using the U-Net architecture, the pancreas is segmented from CT slices in the first stage, while tumors are segmented from masked CT images in the second stage. For validation set of NIH dataset and according to the proposed method's dice scores, the pancreas segmentation and tumor segmentation performance was outstanding, demonstrating its potential to identify pancreatic cancer efficiently and accurately.