Predicting Neurologic Disease Progression Using Machine Learning and Causal Inference

Quantifying communication between different brain regions allows us insight into one of the most complex emergent properties of biological systems. Fundamental properties about cognition, health, and disease can be gleaned from this immensely complicated biologic communication web. One of the best, noninvasive ways for analyzing brain connectivity is functional magnetic resonance imaging (fMRI) - which provides a noninvasive way to look at healthy and diseased individuals' neural communication. Starting with existing means of quantifying neural connectivity, this work performs an in-depth cross-analysis of their capabilities before proposing, testing, and applying novel connectivity measures, including both nonlinear machine learning approaches and prior-knowledge informed causal approaches. These measures are tested in healthy individuals before being applied to clinically-relevant use cases across multiple neurodegenerative diseases. This work finds neuroimaging connectivity biomarkers for the diagnosis of Autism spectrum disorder, the diagnosis of Parkinson's disease, and Parkinson's disease progression. The new measures of connectivity prove invaluable for these diagnostic and prognostic models, providing insights into pathophysiology of the disease and generating clinically-relevant models that meet patient needs.