Autism Pattern Classification Based On Magnetic Resonance Imaging

In recent years,with the advancement of science and technology as well as research development,there was an application research that combine machine learning and disease diagnosis.This comprehensive interdisciplinary study has important value for clinical applications.However,there are only few of pattern classification studies based on magnetic resonance imaging for Autism.In addition,there is not much of researches about the difference between autism and Asperger's syndrome.Therefore,this study attempts to conduct research on the above two types of subjects based on structural and functional magnetic resonance data.Research participants include 37 healthy peoples,28 Autism patients,21 Asperger's syndrome participants.Main work and contribution of this paper are as follows two parts:(1)Based on magnetic resonance imaging technology and pattern classification technology,adopting image data from Autism subjects and healthy subjects to carried pattern classification,and then obtain an effective classification model,which will be a diagnosis reference of Autism.We mainly collected subject's magnetic resonance imaging data to calculate relative index,including ReHo,ALFF,DC,GMV,WMV.Through compare the difference between Autism group and healthy group,extract the corresponding imaging data,and then make a pattern classification by using support vector machine algorithm,so then built up classify model and test the accuracy of classification.We found that accuracy of classify model up to 87.10% in pattern classification training and test between Autism group and healthy group,which means this pattern model will exert positive effect in clinical diagnosis.(2)Calculating the related index of resting magnetic resonance image,and look for potential factor of abnormal behavior in Pervasive developmental disorder,also explore the difference between Autism and Asperger's syndrome.We mainly look for the different brain activity zones between Autism group and healthy group through index statistical analysis,and make a correlation analysis with behavior data,and then explore the potential factor that contribute to the abnormal behavior of Asperger's syndrome,including Wechsler Scale,Autism Diagnostic Interview Scale.We found that there are some significant differences in two sample T test between two groups.For ALFF and ReHo,brain zones that Asperger's group significant higher than healthy group include left angular and right cerebellum,also brain zones that Asperger's group significant lower than Autistic group include frontal orbital cortex and middle frontal cortex.For ALFF,brain zones that Asperger's syndrome group significant lower than both of healthy group and Autism group include right middle frontal cortex.For ALFF of right angular,Asperger's group significant higher than healthy group but lower than Autism group.This dissertation is an important attempt that applying the machine learning to disease diagnosis.What's more,we have a better understanding about Asperger's syndrome,which will lays a solid foundation for the future research work of this thesis.