Research and Application of Medical Image Segmentation Method Based on CNN

Medical image segmentation is an important research topic, among which the convolutional neural network (CNN) has excellent performance in computer vision and pattern recognition. It can extract high-order spatial features from images, categorize pixels based on features, calculate the loss value from the classification results and pixel category labels, and optimize the results to ensure consistency with the category labels. Thus the segmentation of the original image is realized. This paper reviews the current medical image processing system based on convolutional neural networks, introduces its theoretical basis, selects the most popular semantic segmentation model, U-Net network, and adds an attention mechanism to concentrate the most beneficial data in the image. In this paper, the model structure was built by PyTorch, and the simulation experiment was carried out to realize the hip joint segmentation.