Computer Aided Decoding of Brain-Immune Interactions in Gulf War Illness (GWI): A Joint Embedding on Brain Connectomic and Immunogenetic Markers

Brain-immune interaction is a key factor to understanding the origin of complex symptoms in GWI. This project was designed to study the brain-immune interactions by utilizing a novel computer aided decoding scheme. Throughout 4 years of project period, we developed an image processing pipeline and computational frameworks for analyzing multimodal magnetic resonance imaging (MRI) data collected from the Gulf War Illness Consortium (GWIC). Based on this, we have identified potential diagnostic markers for GWI from an abundant amount of imaging markers and also built a machine learning (ML) framework for classifying GWI in a single-subject level. Selected MRI biomarkers were cross-compared to following markers: cognitive, blood immune, and central immune markers. We also examined exposure/symptom-specific MRI markers. ML classifiers were tested across a single modality MRI, multi-modality MRI or MRI markers combined with others (e.g., cognitive or blood immune markers) to identify the best performing models for diagnosing GWI. From these works, key findings include, 1) Diffusion MRI markers provided best performing ML model, 2) Joint embedding of MRI and non-imaging markers help enhancing the ML performance, 3) More clear brain immune interactions found in exposure-specific subgroups, 4) There are variant MRI markers associated with different symptoms. We published these findings in scientific journals and the results were also presented in scientific conferences. Findings (i.e., features and statistical results) and techniques developed in this project were also packed into a software package as planned and ready to be shared to the GWI research community through collaborations to support building more practical and robust technical as well as clinical solutions for GWI.