SMRI image classification method based on high-resolution complementary attention UNet classifier

Abstract

The invention relates to an sMRI image classification method based on a high-resolution complementary attention UNet classifier, and the method is characterized in that the method comprises the steps: obtaining a certain number of sMRI images and labels thereof, carrying out the preprocessing of all sMRI images, and forming a sample set; dividing the sample set into a training set, a verification set and a test set; a network model is constructed and trained and verified, the constructed network model is formed by inserting a feature fusion network and a classification network into an existing Unet network, and the designed feature fusion network is used for supplementing semantic information for an encoder of the Unet network and supplementing detail information for a decoder; the classification network aims to effectively fuse multi-semantic feature maps, so that the expression ability of the network is improved, and classification is realized; and finally, inputting the to-be-tested images in the test set into the optimal network model to obtain a classification result of the to-be-tested images. Therefore, the classification method is simple, and the classification accuracy is improved.

Description

technical field [0001] The invention relates to the field of image processing, in particular to an sMRI image classification method based on a high-resolution complementary attention UNet classifier. Background technique [0002] Traditional classification methods for structural magnetic resonance imaging (sMRI) images are limited by the complexity of manually designed feature extraction and the risk of potential feature loss. In recent years, many image classification methods based on convolutional neural network (CNN) have powerful task-oriented feature representation capabilities, and these CNN-based methods can be divided into four categories: 1) slice-based methods, 2) The method based on image blocks, 3) the method based on regions-of-interest (ROI for short), and 4) the method based on the whole image. [0003] The slice-based method is to use the two-dimensional slices extracted from the original three-dimensional image as the input of the two-dimensional CNN. This ...

AI Technical Summary

Problems solved by technology

However, when fusing shallow information, not all shallow information has a positive effect on image recognition, on the contrary, too much useless shallow information is not conducive to the correct classification of images

Images