Deep learning system for abdominal disease detection and localization

CT images are usually grayscale images and contain a large amount of medical information. However, deep learning research is still relatively insufficient for automatically detecting and segmenting lesions in abdominal CT images. This is due to the complexity of abdominal organs and their variability among different patients, which poses a high challenge to the algorithm. Using the improved Unet model combined with the transfer learning method for abdominal CT image segmentation can significantly improve the detection and diagnosis accuracy of liver tumors, kidney tumors, appendicitis, and other lesions. The Unet model was originally designed for the segmentation of biomedical images. It has end-to-end learning capabilities and can perform fine segmentation at the pixel level. The improved Unet model has made some important improvements based on the original structure, including adding data preprocessing and attention mechanisms, thereby improving the model’s segmentation accuracy and generalization ability. The transfer learning method uses a pre-trained model on a large data set and then applies it to the task of abdominal CT image segmentation. To validate our proposed approach, we conducted rigorous testing on a dedicated dataset. We evaluated key metrics such as accuracy, specificity, sensitivity, and DICE coefficient to evaluate its effectiveness.