A GALLBLADDER CANCER CLASSIFICATION MODEL BASED ON GLOBAL-LOCAL NET

Medical imaging data-based computer-aided diagnosis technology is being employed more and more frequently due to the quick advancement of artificial intelligence technology. The most frequent malignant tumor in the biliary tract, gallbladder cancer has a high degree of aggressiveness and a bad prognosis. As a benign chronic gallbladder inflammation, xanthogranulomatous cholecystitis is non-specific in clinical manifestation compared with gallbladder cancer. Preoperative correct identification of gallbladder cancer and xanthogranulomatous cholecystitis is of great importance for doctors to choose surgical methods. Therefore, this paper intends to establish a classification model for gallbladder cancer and xanthogranulomatous cholecystitis through deep learning methods to provide clinical decision support for doctors. This paper analyzes the performance of convolutional neural network and Transformer algorithms in terms of global feature and local feature extraction, and proposes Global-Local Net. Using CT images as datasets, classification models for gallbladder cancer and xanthogranulomatous cholecystitis were constructed based on EfficientNet, Vision Transformer and Global-Local Net, respectively. According to the testing findings, the accuracy of the Global-Local Net can reach 0.9000 and 0.943, and the AUC can reach 0.848 and 0.895 on the arterial phase CT dataset and venous phase CT dataset, respectively.