Semi-supervised medical image classification method based on security comparison self-integration framework

Abstract

The invention provides a semi-supervised medical image classification method based on a safety comparison self-integration framework, and the method comprises the following steps: firstly, selecting a batch of data composed of labeled and unlabeled data, adding data disturbance to the data, repeating the processing twice to obtain two groups of data, and inputting the two groups of data into a student network and a teacher network respectively; secondly, designing a weighting function, updating the weighting function by utilizing supervised loss, and automatically distributing a weight for each piece of unmarked data; the weight parameters and the probability output of the two networks are combined, and the consistency loss of reliable perception is established; further, normalized low-dimensional representations output by the two networks are obtained by combining weight parameters and utilizing a projection network, and comparison loss of reliable perception is established; and finally, respectively carrying out weighted summation on all the loss functions to form a final loss function, and alternately updating network parameters and weighting function parameters. According to the method provided by the invention, reliable data level and data structure level information can be learned at the same time, and the robustness and generalization of the model are improved.

Description

technical field [0001] The invention relates to a semi-supervised medical image classification method based on a safety comparison self-integration framework, which belongs to the field of medical image classification. Background technique [0002] In recent years, deep learning has made great breakthroughs in the field of clinical medicine and is widely used in pathology, medical imaging and diagnosis, bioinformatics, etc. Generally, deep learning requires a large amount of labeled data for network learning. The time-consuming and labor-intensive acquisition of large amounts of high-quality labeled medical data and the need for specialized medical knowledge hinder the demonstration application of deep learning methods in clinical practice. Deep semi-supervised learning methods can utilize a large amount of unlabeled medical data to assist in the establishment of deep networks, among which, self-ensemble models have been shown to achieve better semi-supervised medical image...

AI Technical Summary

Problems solved by technology

However, the current self-integration models ignore the reliability problem of unlabeled medical images

There are inevitably differences between unlabeled medical images obtained from different populations, devices, and environments. Using unlabeled medical images directly and assigning the same weight to all unlabeled data may affect the classification performance of self-integration methods. Negative impact, existing methods always have limitations in terms of security and generalization in practical applications

Images