Abstract:
Epilepsy is one of the major neurobiological disorders and it can induce asymmetries of
brain regional activation and connectivity patterns in epileptic patients as compared to healthy
controls. These asymmetries can discriminate epileptic patients from healthy controls. This needs
robust biomarkers that could find these asymmetries that can be achieved with the help
functional magnetic resonance imaging (fMRI) data due to its high spatial resolution. In this
thesis, we present two novel biomarkers, Dissimilarity of Activity (DoA) and inter subject blood
oxygenated level dependent (BOLD) signal asymmetries (iBSA) that can capture asymmetries of
activity by finding abnormalities in BOLD signal. We used functional connectivity analysis to
find underlying connectivity patterns of regions among each other. Combining DoA and iBSA
biomarkers with functional connectivity analysis of regions yields a feature set that can represent
all asymmetries of connectivity and activation pattern of regions. We used these features to
discriminate healthy controls from epileptic patients with the help of Support Vector Machines
(SVMs). We achieved classification accuracy of 87.4% on 50-50 split for training and testing
data. Results achieved by our proposed classification model are better than any other
classification model on this dataset in current literature.
