Medical image classification method and device based on multi-scale perception loss

Abstract

The invention relates to a medical image classification method and device based on multi-scale perception loss, and the method employs a convolutional neural network to process a medical image, and also comprises the steps of transmitting the output of the convolutional neural network to at least two class activation graphs and a classifier; transmitting the outputs of the at least two class activation graphs to a first loss function for further mining a region of interest in the outputs of the at least two class activation graphs; transmitting the output of the classifier to a second loss function used for increasing the weight of the concerned area and reducing the weight of a non-concerned area; and fusing the output of the first loss function and the output of the second loss function together to obtain a classified image. According to the present invention, the first loss function can promote the model to more accurately pay attention to a pathogenic area; the second loss function is helpful for increasing the loss weight of the difficult-to-recognize diseases and reducing the loss weight of the easy-to-recognize diseases, so that the classification of the convolutional network on the difficult-to-recognize samples is enhanced.

Description

technical field [0001] The present invention relates to the field of image classification, in particular to a medical image classification method based on multi-scale perceptual loss and a medical image classification device based on multi-scale perceptual loss. Background technique [0002] Classification is to find an accurate description method for each class according to the class characteristics exhibited by the data in the training data set, thereby generating a class description or model, and using it to classify new data. And this type of description is the filtering, extraction and concept of source data, which can reflect the general laws of objective things. Medical image data classification is based on the image image sample data of historical clinically confirmed cases, combined with expert knowledge to form a class description, and based on this, it can classify and predict clinical image images of unknown categories, and assist doctors in clinical image analys...

AI Technical Summary

Problems solved by technology

Although the existing methods have achieved great success, in the application of medical images, the corresponding part of the visual presentation of the multi-label classification model before and after the image transformation is inconsistent. How to make the model more accurately focus on the pathogenic region, There is no good solution to the problem of sample positive and negative category imbalance.

Images