| Summary: | This study explores the use of network analysis on resting-state functional MRI (fMRI) data to
identify distinct brain connectivity patterns in individuals with Autism Spectrum Disorder (ASD)
compared to healthy control subjects, and to be able to distinguish them. The Brain Imaging Data
Exchange (ABIDE) dataset was used, comprising of a similar number of ASD and control
subjects. Within the fMRI scans, the brain regions were modeled as nodes, and their interactions
as edges. A data preprocessing pipeline was used to extract time series data from the Regions of
Interest (ROIs) of the brain defined by a parcellation atlas. Functional connectivity matrices were
then derived from fMRI scans, whereby graph metrics such as degree centrality and clustering
coefficient were computed and visualized. Multiple machine learning classifiers, such as a
Support Vector Machine (SVM), a Graph Neural Network(GNN) and a Convolutional Neural
Network(CNN) and more, were then used to distinguish between the two groups based on their
brain connectivity patterns, and to classify subjects based on their diagnostic labels. The models
achieved varying results and classification accuracy, with certain models performing better than
the others, revealing differences in long-range connectivity between frontal and parietal regions
in ASD subjects. These findings highlight the potential of graph-based methods in neuroimaging
research and for understanding the neural mechanisms of ASD. Furthermore, this study provide a
foundation for future work in early diagnosis and understanding of neurological disorders.
|
|---|