A Novel Coarse-to-Fine Segmentation Method for Pancreatic Cancer

Pancreatic cancer is one of the most deadly cancers, with a 5-year survival rate of only 8%, but early diagnosis can increase the survival rate to about 20 /0. Therefore, by examining abdominal CT scans, early and accurate diagnosis of pancreatic tumors is critical to treatment decisions. In CT images, pancreatic cancer faces imbalanced classification, background interference, and non-rigid geometric features, making it difficult for even experienced radiologists to diagnose. Therefore, we propose a novel method to detect pancreatic cancer more accurately. In order to make the positioning more accurate, we use a rough segmentation network to get a rough positioning, and then get a more accurate segmentation effect based on this positioning. We trained and tested on a public data set containing 281 cases, and the dsc coefficient reached 63.4 +/- 23.67%. This experiment proves that our proposed new method is effective and shows potential clinical impact.