Investigation of brain connectivity in autism spectrum disorder using sparse inverse covariance method and SVM classifier

Autism Spectrum Disorder (ASD) is a developmental disorder that affects social communication and behaviour. Many functional neuroimaging studies helped to establish that ASD is a neurological disorder. Even with this knowledge, brain biomarkers are unknown, and diagnosis is still behaviourally ba...

وصف كامل

محفوظ في:
التفاصيل البيبلوغرافية
المؤلف الرئيسي: Vineetha, Koneru
مؤلفون آخرون: Justin Dauwels
التنسيق: Theses and Dissertations
اللغة:English
منشور في: 2019
الموضوعات:
الوصول للمادة أونلاين:http://hdl.handle.net/10356/77963
الوسوم: إضافة وسم
لا توجد وسوم, كن أول من يضع وسما على هذه التسجيلة!
المؤسسة: Nanyang Technological University
اللغة: English
الوصف
الملخص:Autism Spectrum Disorder (ASD) is a developmental disorder that affects social communication and behaviour. Many functional neuroimaging studies helped to establish that ASD is a neurological disorder. Even with this knowledge, brain biomarkers are unknown, and diagnosis is still behaviourally based. In the past few years, diagnostic classification of neurological disorder patients and neurotypical patients using functional connectivity has drawn great interest and shown some significant results. To study functional connectivity, Sparse Inverse Covariance Estimation (SICE) is known as an effective tool. Hence in this study, a SICE method has been implemented to identify the functional connections between brain regions. Pattern recognition methods are then applied to identify discriminatory features and train the machine to identify ASD patients in early developmental stages. Resting-state functional magnetic resonance imaging scans of 600 typically developing (TD) and ASD participants, matched for age, gender and root mean square deviation have been selected from Autism Brain Imaging Data Exchange. For a threshold value of 0.1 SICE matrices are calculated using Max-Det Matrix Completion (MDMC) method. Features were selected based on random forests’ mean decrease in accuracy measure. Most elevated accuracies are acquired prevalently for poly and radial kernels. Maximum accuracy of 72.5% is achieved utilizing poly kernel for 857 features with 2500 trees. On average 68.5% accuracy is achieved.