Resting-state EEG and machine learning to investigate cortical connectivity as a biomarker in chronic mTBI

Mild traumatic brain injury (mTBI) is a heterogeneous condition with long-term sequelae, yet diagnosis in the chronic stage remains limited by reliance on acute criteria and subjective reports. Objective biomarkers are needed, as current blood-based markers show diagnostic value primarily in the acute and subacute phases. Resting-state EEG (RS-EEG) can capture large-scale network disruptions through functional connectivity (FC) and microstate analysis, but its role in chronic mTBI is unclear. We tested whether RS-EEG features distinguish chronic mTBI from controls and predict symptom burden. This observational case-control study included 44 participants (18 chronic mTBI, 26 controls). Source-reconstructed EEG was analyzed for spectral power, microstate metrics, and FC using the Multivariate Interaction Measure (MIM). Elastic Net and XGBoost models classified injury status and predicted symptom severity, with feature robustness evaluated across full and reduced electrode montages. Participants with mTBI showed no group differences in spectral power or microstate metrics but demonstrated significantly elevated FC across theta, beta, gamma, and broadband frequencies. Connectivity increases were stable across reduced montages and persisted up to eight years post-injury. Classification models using MIM achieved AUCs of 0.79–0.89 for injury status and 0.82–0.87 for symptom severity, outperforming demographic models. Resting-state EEG FC provides a sensitive biomarker of chronic mTBI, distinguishing cases from controls and correlating with symptom severity. The persistence of network alterations years after injury suggests lasting changes in brain activity associated with chronic symptom burden. These findings support the use of RS-EEG–derived FC as a noninvasive and scalable biomarker of chronic mTBI.