BRAINGNNet: Graph Learning Models for Chronic Pain Detection through EEG Biomarkers

Advancements in healthcare have yet to resolve the longstanding challenges associated with the objectiveassessment of pain, particularly chronic pain, due to the complexity of identifying specific neuro markers. Clinical practices still predominantly rely on subjective assessments and rating scales, requiring more accuracy, reproducibility, and broader applicability for widespread clinical or research adoption. This study explores a novel approach to chronic pain detection leveraging Electroencephalographic (EEG) biomarkers. As part of our discovery phase, we developed BRAINGNNet, a graph-based model integrating brain topology with EEG channel data and extracted features, utilizing open-source tools and publicly available datasets. The feasibility of this method was evaluated through a proof-of-concept implementation of a Heterogeneous Graph Attention Network (HAN). We successfully distinguished chronic pain states from control conditions by extending message-passing and attention mechanisms with meta-paths emphasizing pain pathways. Our preliminary findings, supported by BNNHan’s promising predictive performance, demonstrate the potential of this lightweight, data-driven framework to enhance chronic pain detection, monitoring, and management in clinical applications. These results pave the way for more scalable and resource-efficient solutions to address chronic pain as a critical healthcare challenge.