Nuclear magnetic resonance image small sample learning classification method based on graph network

Abstract

The invention relates to a nuclear magnetic resonance image small sample learning classification method based on a graph network, and the method comprises the following steps: 1, inputting a disease original image sample into a feature extraction part network in an original nuclear magnetic resonance image small sample learning classifier, and obtaining a corresponding feature vector; 2, inputting the feature vectors into a graph network in an original nuclear magnetic resonance image small sample learning classifier, initializing nodes and edges, and carrying out continuous iterative updating; 3, after continuous iteration updating of the nodes and the edges is completed, obtaining a nuclear magnetic resonance image small sample learning classifier based on the graph network; and 4, performing prediction classification on the actual nuclear magnetic resonance image by using the new classifier. Compared with the prior art, the invention has the advantages that the classifier capable of accurately judging the staging condition of the disease corresponding to the image can be quickly trained by learning a small number of samples in the face of a new disease, so that the problem of lack of labeled samples faced by medical image classification is solved, and the training cost is reduced.

Description

technical field [0001] The invention relates to the technical field of neural networks, in particular to a method for learning and classifying small samples of nuclear magnetic resonance images based on graph networks. Background technique [0002] The analysis of Magnetic Resonance Imaging (MRI) plays an important role in the discovery and treatment of patients' diseases. Applying deep learning methods to analyze MRI is a typical scenario in the context of artificial intelligence empowering the medical field. The mainstream method currently used is 3D convolutional neural network and its variants. This type of method automatically extracts the deep features in the data by selecting the appropriate number of layers and model parameters, and then classifies the MRI images. In order to improve its classification performance, it is necessary to enhance its feature extraction ability and take into account its generalization ability. Usually, improvements can be made by modifyi...

AI Technical Summary

Problems solved by technology

In the medical field, the number of high-quality image samples is scarce, the cost of labeling is high, and the ratio of the number of control samples to case samples is extremely unbalanced, which makes it difficult for the classifier to obtain sufficient labeled samples and the accuracy of the model needs to be improved.

Images