Please use this identifier to cite or link to this item:
|Title:||Methods for accurate diagnosis and insights of Autism Spectrum Disorder using functional and structural magnetic resonance imaging||Authors:||Subbaraju, Vigneshwaran||Keywords:||DRNTU::Engineering::Computer science and engineering
|Issue Date:||2016||Abstract:||Autism Spectrum Disorder (ASD) refers to a cluster of relatively common developmental disorders characterized by impairments in social behavior and communication. Accurate diagnosis of ASD is crucial for providing customized remedial therapy. Conventional interview based methods of clinical diagnosis are error prone and they are also unable to point out any biological basis behind the observed behavioral symptoms. Brain imaging is considered as an alternative to overcome these difficulties. Among the brain imaging modalities, structural MRI (sMRI) provides the anatomical details of the brain at a high spatial resolution while functional MRI (fMRI) provides accurate details of connectivity, which helps in understanding the changes in the brain due to ASD. However, several factors such as age, gender etc. influence the development of the brain under ASD. To enable detailed studies of ASD in the presence of such factors, a comprehensive dataset of MRI scans, obtained from people of various age and gender was released for public use by the Autism Brain Imaging Data Exchange (ABIDE) consortium. In this thesis, the differences in functional connectivity due to ASD are first analyzed using the resting-state fMRI data provided by ABIDE. For this purpose, the Blood Oxygen Level Dependent (BOLD) signals belonging to 90 regions of the brain (using Automatic Anatomical Labeling (AAL) template) are obtained as a time-series from the pre-processed fMRI scans. A new method called Spatial Feature based detection Method (SFM) is proposed in this thesis, which uses the BOLD time-series signals to identify the variations in functional connectivity and achieve accurate diagnosis of ASD. In SFM, a spatial filter is determined which can project the covariance matrices of ASD patients and neurotypical subjects in orthogonal directions. The projected time-series obtained using the spatial filter correspond to the discriminative neural activity that offer maximum separability between the two classes in terms of variance. From this, the most dominant log-variance features are extracted for accurate diagnosis of ASD. The spatial filter coefficients represent the influence of each region over the others for accurate diagnosis. Using this spatial filter, SFM also provides a spatial pattern map within the brain, which corresponds to the differences in functional connectivity between the ASD patients and neurotypical subjects. With reference to medical literature, separate gender and age-group specific studies are conducted in this thesis, to highlight the specific pattern variations for each category. For adolescent males, the maps show that the posterior cingulate gyrus, transverse temporal gyrus, thalamus and some parts of pre-frontal cortex are the regions where significant differences are observed in resting state activity between ASD patients and neurotypical subjects. For adult males, the amygdala, postcentral gyrus and calcarine sulcus are affected in addition to the regions identified for adolescent males. For adolescent females, parts of the prefrontal cortex, insula, anterior/posterior cingulate gyrus are the regions where significant differences in resting state activity are observed. For adult females, anterior cingulate gyrus, parahippocampal gyrus, fusiform gyrus, calcarine sulcus, cuneus and paracentral lobule regions are affected by ASD. Using the log-variance features provided by SFM and a gender and age-group specific classification approach, better ASD detection accuracy is achieved in this thesis (86.5% for adolescent males, >99% for adult males/females and adolescent females) compared to the existing methods in literature (60 to 87%) which directly use the pairwise correlation co-efficients as the features. The spatial pattern map also provides a new insight on connectivity variations which cannot be found by just observing the pairwise correlation co-efficients alone, as done by the previous studies in the literature. In addition to the log-variance features, this thesis also proposes an enhanced effect size thresholding method for accurate diagnosis of ASD from fMRI. When compared to enhanced effect-size features and other methods, the log-variance features obtained from SFM provides higher classification performance for diagnosis of ASD. The anatomical variations in the brain which can be observed using sMRI are then studied for ASD diagnosis. Among the components of the brain, gray matter plays an important role in cognitive development. In this thesis, automatic, whole brain, Voxel Based Morphometry (VBM) analysis of sMRI is used to identify the differences in gray matter and detect ASD. Unlike earlier studies in the literature, this thesis provides a comprehensive gender and age-group specific analysis using the entire ABIDE dataset. The studies show that the premotor cortex and supplementary motor cortex are affected for adult-females while the somatosensory cortex is affected for adolescent-females with ASD. For adolescent males, the precentral gyrus, motor cortex, medial frontal gyrus and the paracentral lobule areas are affected while the superior frontal gyrus and the frontal eye fields areas are affected for adult males due to ASD. An important challenge in using whole brain VBM analysis for ASD diagnosis is the high dimensionality of the features obtained. To handle this, an Extended Metacognitive Radial Basis Function Neural Classifier (EMcRBFN), which is an improved q-Gaussian RBF neural network classifier has been developed in this thesis. Using the EMcRBFN classifier, classification accuracies of 98.75% and 85.86% are obtained for adult females and adolescent females respectively, and accuracies of 61.49% and 70.41% are obtained for adolescent males and adult males respectively. These are about 5 to 11 percentage points higher than the accuracy obtained using SVM. The classification results from both sMRI and fMRI show that accurate ASD detection is more challenging for males compared to females. Within each gender, it is more difficult to diagnose adolescents whose brains may inherently show higher natural variations since they undergo rapid development following a unique trajectory, compared to the adults whose developed brains may show lesser natural variations. This study validates the findings in medical literature, which have observed gender and age-group specific effects of ASD and clearly shows that the different categories must be considered separately to obtain accurate diagnosis and biological interpretations from MRI.||URI:||http://hdl.handle.net/10356/68823||Fulltext Permission:||restricted||Fulltext Availability:||With Fulltext|
|Appears in Collections:||SCSE Theses|
Updated on May 12, 2021
Updated on May 12, 2021
Items in DR-NTU are protected by copyright, with all rights reserved, unless otherwise indicated.