Using Electroencephalography-Based Connectomics for Pattern Learning on Psychiatric Disorders

Electroencephalography (EEG) based functional connectivity maps the interaction between brain regions with high temporal resolution. In addition, this functional connectome can separate complex neural interactions into distinct frequency bands. This study constructed a pattern learning framework to explore the brain network topology of healthy individuals and patients with psychiatric disorder. Graph theory was applied to the time-averaged resting state networks of social anxiety disorder patients and veterans suffering from post-traumatic stress disorders. Similar analyses were repeated with connectomes of healthy controls during the task performance. A manifold learning model was applied to the temporally dependent EEG network to explore network dynamics during emotion regulation. The graph analyses identified altered network properties in both social anxiety disorders patients and combat-exposed veterans with post-traumatic stress disorders. Theta connectivity is associated with emotion regulation effectiveness. Properties extracted from the constructed manifold phase-space reflect the cognitive load of the task and disease diagnosis. Selected prototypes of the phase-space serve as predictors to classify task condition and diagnose in a nearest neighbor informed supervised learning. This framework provides a novel approach to understand the underlying neurophysiological abnormality of psychiatric disorders. The identified patterns and frequencies of interest can serve as useful biomarkers to facilitate in disease diagnosis, treatment progression and task performance evaluation.