Functional Network Connectivity(FNC)-based Generative Adversarial Network (GAN) And Its Applications In Classification Of Mental Disorders

Mental disorders cause high socioeconomic burdens and many disease exhibit comorbidity between each other.However,diagnosis of mental disorders mainly depends on symptom scores from clinical interview,which is subjective and has a strong dependence on the professional quality of doctors.Therefore,finding the potential biomarker for patients with mental disorders diagnosis has become a hot topic of common concern all over the world.Functional network connectivity(FNC)has been commonly utilized to study mental disorders,which can reflect the organization and interrelationship of spatially separated brain regions.FNC is widely applied in neuroimaging to identify potential biomarkers for predicting or classifying mental disorders.As a popular deep learning method,generative adversarial networks(GAN)have achieved outstanding performan ce in multiple classifications and segmentation tasks.However,the application of GANs to fMRI data is relatively rare and,the high-dimensional data in fMRI field is often with only a limited sample size,which makes it challenging for the classifiers to learn a good decision boundary.In this work,for the problem of high-dimensional small samples of mental illness data we proposed a functional network connectivity(FNC)based GAN for classifying psychotic disorders from healthy controls(HCs)and achieved the improvement the classification accuracy.The main achievements of the thesis are as follows:In this work,we developed a novel GAN model using resting-state fMRI to discriminate mental disorders from HCs in large,multi-site datasets.Our model alleviates the small size problem of FNC images by making full use of generated FNC samples.In addition,the proposed GAN model defines adversarial objections between the generators and discriminator,which uses adversarial learning and feature matching to further improve the classification performance of the discriminator.The proposed GAN model consisted of one discriminator(real FNCs)and one generator(fake FNCs),each has four fully-connected layers.The generator was trained to match the discriminator in the intermediate layers while simultaneously a new objective loss was determined for the generator to improve the whole classification performance.In a case for classifying 558 schizophrenia patients from 542 HCs from 7 sites,the proposed GAN model achieved 80.7% accuracy in leave-one-site-out prediction,outperforming support vector machine(SVM)and deep neural net(DNN)by 3%-6%.In another application to discriminating 269 major depressive disorder(MDD)patients from 286 HCs,an average accuracy of 70.1% was achieved in 10-fold cross-validation,with at least 6% higher compared to the other 6 popular classification approaches(54.5-64.2%).More importantly,we are able to identify the most contributing FNC nodes and edges with the strategy of leave-one-FNC-out recursively.To the best of our knowledge,this is the first attempt to apply the GAN model on the FNC-based classification of mental disorders.Such a framework promises wide utility and great potential in neuroimaging biomarker identification.