Medical Image Classification Based on Convolutional Neural Network

This study proposes a medical image classification method based on convolutional neural network (CNN), ResNet and Attention Mechanism (AM). A new Convolutional Residual Attention Network (CRAN) model is constructed. Initially, CNN is utilized to extract localized features from medical images and capture the intricate details within them. Secondly, the ResNet residual connection is used to tackle the problem of vanishing gradients in deep networks and enhance the model's expressiveness. Ultimately, the introduction of the attention mechanism allows for the weighting of the feature map and dynamically directs focus towards significant areas within the image, thereby enhancing classification accuracy. In the experiment, we compared with traditional methods, and the results showed that the classification accuracy of CRAN model on multiple medical image data sets was significantly higher than that of traditional methods, especially when dealing with complex lesion areas, the advantages of CRAN model were more obvious. This result fully validates the feasibility and effectiveness of CRAN architecture in medical image classification, and provides a new idea and method for medical image classification.