Deciphering brain heterogeneity in neuropsychiatric disorders with artificial neural networks

Abstract

In recent years neuropsychiatric disorders have emerged as leading causes of disability and mortality worldwide. While these conditions are multifaceted, they share common characteristics such as heterogeneous and overlapping symptomatologies and poorly understood etiologies, giving them an enigmatic appearance. One way to elucidate their biological underpinnings is through Magnetic Resonance Imaging (MRI), an imaging modality that produces high-resolution images of the brain non-invasively. However, analyzing MRI images has so far been unable to reveal canonical patterns of aberrations distinctive to these disorders. The renaissance of Artificial Intelligence (AI) and deep learning operationalized through deep, artificial neural networks have over the last decade yielded monumental progress in image processing, where machines learn to detect and leverage patterns in imaging data to reach predictive tasks. AI could present a pivotal opportunity to use technology to further our understanding of the biological aberrations underlying neuropsychiatric disorders. In the current thesis, we trained multiple deep neural networks to predict various outcomes based on structural MRI scans of the brain. Next, we used both the predictions from the models and the knowledge they had learned to characterize heterogeneity in neuropsychiatric patients. We show that this technology has the potential to help us elucidate the biology underlying the disorders, and procure tools that can support precise and personalized clinical decision-making.