Adaptive medical image classification method and system based on Transform

Abstract

The invention provides a Transform-based adaptive medical image classification method and system, and the method comprises the steps: obtaining basic image feature information through a convolutional neural network, transmitting the image feature information to a channel attention network structure, obtaining a weighted channel attention feature map, transmitting the channel attention feature map to a Transform network, and obtaining a weighted channel attention feature map; and finally, inputting the obtained feature vector into a full connection layer, and directly outputting a final classification result of the algorithm by the full connection layer. According to the method, the convolutional neural network and the Transform model are effectively combined together, so that the redundancy of the model is reduced, and the global context information of the image can be captured to extract powerful feature information.

Description

Technical field [0001] The present invention relates to the field of imaging classification technologies based on artificial intelligence, and Background technique [0002] With the continuous accumulation of medical imaging, a huge challenge has brought a huge challenge to a medical imaging of medical imaging to use a traditional machine learning model. The depth learning algorithm has matured in the computer visual field, providing an opportunity for the auxiliary doctor to perform accurate diagnosis. An important branch of convolutional neural networks as deep learning is a very important location in image processing, and the characteristics of local perception and parameter sharing of convolutional neural networks can effectively extract image features and reduce model complexity. Currently classic convolutional neural network models such as the VGG model, the RESNET model, and the Densnet model are widely used in the characteristic extraction of medical imaging. The prior ar...

AI Technical Summary

Problems solved by technology

Although the convolutional neural network has so many advantages, due to the local perception and parameter sharing characteristics of the convolutional neural network, there is less correlation information between the pixels of the image, which is not conducive to the network's full use of context information for feature capture.

Although the convolutional neural network can extract the feature receptive field covering the whole image by continuously superimposing deeper network layers, this will obviously increase the amount of network calculation, causing network training problems such as gradient dispersion.

Images