Research On HIV Brain Image Data Enhancement Algorithm Based On Generative Adersaral Networks

Since artificial intelligence technology was proposed,it has been widely used in many fields and achieved remarkable success.Although deep learning model can surpass human performance in multiple tasks through training,its excellent performance often depends on massive training data.However,in the field of artificial intelligence assisted medical research for special diseases such as HIV,due to the high cost of data collection,medical images involving patient privacy and other issues,there is a lack of medical data for research,which will seriously affect the training effect of the follow-up model.Therefore,it is necessary to study the algorithm that can be used for medical image data enhancement.In the early stage of HIV infection,the virus infects macrophages and enters the nerve center through the blood-brain barrier,causing chronic long-term damage of nerve cells.Even if antiretroviral therapy is carried out at this stage,nearly half of HIV infected people still have neurocognitive and operational dysfunction,namely HIV-associated Neurocognitive Disorders(HAND).There will be asymptomatic neurocognitive impairment(ANI)in the preclinical stage of HAND,this stage has no clinical symptoms or mild symptoms,but it is the earliest stage of HAND neuropathy.If we can accurately predict the diagnosis and carry out targeted intervention in this stage,it will have important scientific significance and clinical value to improve the prognosis and quality of life of hand patients.This paper focuses on the brain medical image data of patients with ANI,based on the generative adversarial network model,studied and proposed an algorithm to generate high-quality medical synthetic images and enhance the original hand medical image dataset.Aiming at the shortcomings of the original WGAN network model,which does not effectively use the hidden space features of real samples and the network structure is too simple,the original WGAN network model is improved.An encoder network is added before the generator network of the original model,and the hidden space features of the real samples are extracted and used to assist the generation of generator network training,the channel attention mechanism is introduced into the encoder network to improve the network learning ability.At the same time,inspired by the residual block structure in the deep residual network,the residual block structure is added to the generator network of the original model to further improve the network learning ability.Experiments show that the improvement of the original WGAN model can effectively improve the quality of the generated image,but there is still a gap in the definition between the generated image and the real image.In order to further improve the quality of the generated image,the edge detection strategy is introduced,the edge texture loss is added to the the generator network of the improved model,so that the generator network can learn more detailed feature information.The experimental results show that the edge detection strategy can play a positive role in improving the quality of the generated image.At the same time,in order to more comprehensively prove the effectiveness and usability of the image generated by the improved model,the image generated by the model is applied to the classification task,the image generated by the model is used to assist the real image to complete the two classification task of ANI/NC.The classification results show that the image generated by the model can significantly improve the accuracy of the classification model.