A screening device and method for diabetic retinopathy based on adversarial learning

Abstract

The screening device for diabetic retinal disease based on semi-supervised adversarial learning of the present invention introduces a second classifier unit, a reconstruction unit and a discriminant network module on the basis of a classification deep neural network (ResNet34), and introduces a semi-supervised adversarial learning mechanism into the entire device. During the training process, the device can be trained using a limited number of annotated and a large number of unlabeled fundus images, which has good practicability and versatility; the introduction of image reconstruction branch, adversarial learning and a second classifier is verified through comparative experiments. The unit to a common classification network can greatly improve the accuracy of its screening of diabetic retinopathy; the device is applied to two datasets, EyePACS and Messidor, and both have achieved good experimental results. The screening device of the present invention can effectively perform the screening of diabetic retinopathy of color fundus images, and plays the role of assisting doctors in making relevant medical diagnosis.

Description

technical field [0001] The invention belongs to the field of medical image processing, in particular to a screening device and method for diabetic retinopathy based on semi-supervised confrontation learning. Background technique [0002] Diabetic retinopathy (DR) is the first blinding eye disease among working-age people, and it is a series of diseases caused by retinal microvascular damage caused by diabetes. In fact, early diagnosis and timely treatment can effectively prevent blindness caused by DR. [0003] In recent years, deep neural networks have been widely used in computer-aided diagnosis systems, and have achieved very promising results in many medical diagnosis tasks. However, training an excellent deep neural network requires a large amount of high-quality annotated medical images. However, since medical image annotation requires expertise and intensive labor, usually only a small number of high-quality annotated images can be obtained in medical image analysis...

AI Technical Summary

Problems solved by technology

[0004] The purpose of the present invention is to overcome the shortcomings in the prior art that a large number of high-quality labeled medical images cannot be obtained to train the deep neural network, and to provide a screening device and method for diabetic retinopathy based on semi-supervised confrontation learning

Images