Schizophrenia, bipolar disorder, major depressive disorder (MDD), and several other psychiatric illnesses share both common and specific clinical behaviors and cognitive impairments. Recent research has provided increasing evidence for a unified framework across these diseases, revealing potential shared neurobiological mechanisms. However, few studies have investigated their common and unique factors using data-driven deep learning algorithms.
This project aims to identify objective neuroimaging biomarkers for these psychiatric disorders by leveraging deep learning techniques, specifically graph neural networks (GNN) combined with spectral harmonic analysis. We will analyze a large, longitudinal dataset from the UK Biobank cohort, which includes multi-modal neuroimaging data (structural, functional, and diffusion MRI), environmental measures, clinical scales, and demographic information. Through this study, we aim to uncover both shared and unique brain activity patterns across these disorders and further identify predictive features for the progression of these diseases. Specifically, GNN will be used to model the complex relationships between brain regions, while spectral harmonic analysis will serve as a useful tool to reveal the temporal and frequency-domain characteristics of brain activity.
The ultimate goal of this research is to identify neuroimaging biomarkers of psychiatric disorders using deep learning algorithms and explore their predictive value in disease progression. Broadly, this study aims to provide new insights the hypothesis of unified framework across psychiatric, offering new directions for early detection and treatment of mental illnesses.