Application Of Graph Neural Network In Auxiliary Diagnosis Of Autism Spectrum Disorder

Throughout the world,autism spectrum disorder(ASD)is recognized as a serious public health problem,and is also considered as a lifelong disease that is unlikely to be cured.ASD patients are mainly characterized by language disorder,social disorder,intelligence disorder and stereotyped behavior pattern,and even adult patients can not live independently,and their abilities in learning,work and other aspects are also relatively poor,which adds a burden to the family and society.However,at present,the way doctors diagnose ASD patients is mostly based on the observation of symptoms and their own experience,and this diagnostic method is very subjective,so it is easy to misdiagnose and miss diagnosis.Moreover,in recent years,the global prevalence of ASD has shown an increasing trend year by year.In this case,it is urgent to develop new methods to assist the clinical diagnosis of ASD.In recent years,with the development of computer technology,the use of deep learning algorithms to assist in diagnosis of diseases has become a research hotspot in the field of mental imaging.As an objective marker,biomarkers can identify nervous system diseases at an early stage,which helps us understand the pathogenesis of potential diseases.Therefore,this paper uses the MRI data obtained from the magnetic resonance scanner to build a model based on the graph convolution neural network algorithm,improve the diagnostic efficiency of the ASD model,and look for biomarkers related to the diagnosis of brain imaging.The specific work content and research results of this paper are as follows:(1)Based on multimodal brain image data,a graph convolution neural network ASD aided diagnosis model is constructed.Using the collected clinical data of three modes of brain imaging,DTI and s MRI fusion data,DTI and f MRI fusion data and DTI,s MRI and f MRI fusion data were selected for research.Through the comparative analysis of the constructed graph convolution neural network aided diagnosis model and support vector machine aided diagnosis model,it was found that DTI and f MRI fusion data had better effect,and its area under the ROC curve(AUC)was 0.78.(2)Based on the graph convolutional neural network algorithm,brain imaging biomarkers related to ASD have been discovered.Due to the high heterogeneity of ASD,this paper selects the nodes that contribute significantly to the model as possible biomarkers through the collected individual level data of brain images,based on the constructed graph convolution neural network model,spectral decomposition,Kolmogorov Smirnov test and other nonparametric tests.The results show that the ASD imaging biomarker may be the prefrontal striatal loop.