GAN-based medical diagnosis model anti-attack method

Abstract

The invention discloses a GAN-based medical diagnosis model anti-attack method for solving the security problem of an artificial intelligence medical image diagnosis model. The method comprises the following steps: building a ResNet-101-based high-precision residual neural network diagnosis model for an acquired medical pathological image, and then building a GAN-based confrontation attack network model which comprises a generator G and a discriminator D, wherein the generator G is used for generating a medical image confrontation sample by superposing high-dimensional random noise disturbance x on an input medical image, and the discriminator D is used for discriminating the authenticity of the confrontation sample; employing a PatchGAN discriminator based on a feature extraction image block for designing three layers of feature blocks including a residual block, expansion convolution and a channel attention mechanism as a main method for feature extraction, so that convolution kernel receptive fields of different scales can extract more refined feature map information by using the method and an effective input medical image disturbance area is obtained; therefore, the anti-attack effectiveness of the medical diagnosis model is improved, and the medical diagnosis model can be reinforced and defended from the anti-attack.

Description

technical field [0001] The present invention relates to the artificial intelligence security field of deep learning, in particular to a GAN-based medical diagnosis model confrontation attack method. Background technique [0002] With the development of AI technology, especially the Deep Learning algorithm, it has recently become an ideal solution choice for smart healthcare. The use of AI medical imaging to assist in the diagnosis of diseases can greatly improve the level and efficiency of disease diagnosis. Although the recognition accuracy and performance of the medical diagnosis system on large-scale medical data sets such as DeepLesion are good, studies have found that adding processed adversarial sample data sets to the trained medical diagnosis model will lead to lethality of the medical model. mistake. [0003] There are two main traditional adversarial attack methods. The first one is a series of variants of FGSM such as FGSM and PGD based on gradient generation. ...

AI Technical Summary

Problems solved by technology

Current deep learning systems usually do not allow white-box access to the model for security reasons, and only allow query access to the model, which treats the model as a black box, and most black-box attacks using traditional adversarial attack methods have a low success rate. high

Images