Autism as a self-helplessness disease is a neurodevelopmental disorder that affects a person's behavior and social interactions. According to the report of the World Health Organization, the preva‎lence of autism spectrum disorder (ASD) is increasing. Currently, the diagnosis of autism is based on clinical interviews and direct observations of behavior. Diagnosis of this disorder only by clinical observations is prone to error. Methods based on magnetic resonance imaging (MRI) have created an exciting way to examine the human brain non-invasively. One of the goals of cognitive science studies in the field of autism spectrum disorder is to help identify this disorder by examining the connections of the people's brain regions. In this regard, researchers have extracted the time series of brain regions by using functional magnetic resonance images in resting state (rs-fMRI) and by calculating the correlation between the time series, they have investigated the functional connections of people's brain. Also, in some researches, they have used diffusion tensor images (DTI) to study the structural connections of brain regions. One of the new perspectives in neuroscience studies is the integration of structural and functional information of brain regions in order to identify ASD. One of the methods that have been invented for this purpose in recent years is the use of signal processing tools on the brain graph (GSP) and in particular graph Fourier Transform (GFT). In most of the studies that have employed GFT to classify subjects ASD from healthly subjects, the actual brain structure of people has not been taken into account, or they have estimated the brain structure from rs-fMRI images, or they have used the structural average matrix of a group of healthy people. In this thesis, by selecting a subset of ABIDE dataset, we have dealt with the problem of classifying people with autism from healthy people. Specifically, by using rs-fMRI and DTI images, we have calculated the GFT signals of 38 brain regions based on the average structural information of healthy people and the structural information specific to each person separately. Then we have obtained the Pearson correlation between the GFT signals of pairs of brain regions. By using the statistical test of analysis of variance (ANOVA), only a number of correlation coefficients have been selected as features and by using the basic component analysis (PCA), the dimensions of the features have been reduced. The data classification was done using support vector machine (SVM), K nearest neighbor algorithm (KNN) and linear discriminant analysis (LDA). The correlation coefficients between the time series of brain regions and structural matrix characteristics have been classified separately. Data classification by calculating the correlation coefficients between the GFT signals based on each individual's specific structural matrix and the average matrix of healthy individuals respectively showed better performance. The results of this research are promising as the first steps in the path of using the structural and functional information of the brain in the identification of autism spectrum disorders

فرزانه شايق، جلال ذهبي

رسول اميرفتاحي، امير اخوان