الفهرس 
Chapter 1 Introduction 1.1 IntroductionOnly 14 pages are availabe for public view
 |  | from | 134 |  |  |
| | | from | 134 | | |
Abstract
Brain disorders have been widely detected in recent years; however, their diagnosis is based on symptom reports performed by patients or profession-als without clinical or quanti ed judgments. Hence, it is prone to human mistakes. There is an urge for an objective computer assisted diagnosing system. One of the challenging brain disorders is the autism spectrum dis-order (ASD) which is a neuro-developmental disorder associated with im-pairments in social and lingual abilities. Failure in language development is variable in the ASD population and follows a wide spectrum. The autism diagnostic observation schedule (ADOS) is the current gold standard for di-agnosing, supported by expert clinical judgment. Early diagnosis allows for early intervention to reduce the severity of autism. Brain scanning technolo-gies have been widely developing and acquired extensively to understand brain functionality and structure. Magnetic resonance imaging (MRI) is a medical scanning technique that uses strong magnetic elds to form images of the anatomy and the physiological functionality of the brain. Main types of MRI include structural, resting-state functional MRI and task-based func-tional MRI (TfMRI). TfMRI demonstrates the functional activation in the brain by measuring blood oxygen level-dependent (BOLD) variations in re-sponse to certain tasks. The aim of this thesis is to develop a personalized computer-aided diagnosis (CAD) and grading system to classify autistic sub-jects against typically developed peers. A novel computer-aided ASD grad-ing framework, dependent on the analysis of brain activation in a response to speech experiment, is proposed. Increased hypoactivation of the superior
iii
iv
temporal cortex, angular gyrus, primary auditory cortex and cingulate gyri is detected with increasing autism spectrum severity. Less lateralization of the left hemisphere regions is also detected. For further local and global feature extraction in the proposed ASD grading system, only the region of interest (ROI) areas are examined. A comprehensive, two stage system is developed using di erent classi ers. Four-fold cross-validation is adopted for testing. The rst stage discriminates between moderate and the other two groups with an average accuracy of 83%. Subsequently, a second stage classi es subjects as mild or severe autism with an average accuracy of 81%. The vali-dation results prove the robustness of the proposed framework for early CAD system to place subjects on the autism spectrum. Recently, deep learning methods have been gaining more attention for fMRI classi cation. However, relatively few studies have applied deep learning techniques to TfMRI for diagnosing autism. For global diagnosis of ASD, a convolutional neural net-work (CNN) based framework and discriminant TfMRI feature extraction techniques are developed. FMRI is considered big data with four dimen-sions. Dimensionality reduction is required to achieve better performance. Therefore, a three-stage pipeline for both temporal and spatial feature ex-traction and reduction is built. Preliminary results on 100 TfMRI dataset (50 ASD, 50 TD) obtain 80% correct global classi cation using 10-fold cross validation. The experimental results show the improved accuracy of the pro-posed framework and hold promise for the presented framework as a helpful adjunct to currently used ASD diagnostic tools. As an early autism local and global CAD tool, A CNN deep local and global ASD classi cation ap-proach with continuous wavelet transform (CWT) is developed. In order to provide a detailed frequency and scale representation, CWT is applied on selected BOLD signals. CWT produces scalograms that provide a detailed representation on these BOLD signals. These scalogram images are used as input images to multi-channel 2D-CNNs for each area. The achieved global accuracy is 86%. Finally, brain maps that indicate level of ASD severity for each ROI are provided for each subject. The proposed framework works towards creating personalized diagnosis and treatment plans that handle the speci c case of each individual.