Abstract Body: Introduction: Predicting consciousness recovery in patients with disorders of consciousness (DoC) after acquired brain injury, such as anoxic brain injury post-cardiac arrest, is critical for treatment decisions. However, uncertainty remains, leading to avoidable morbidity and mortality.
Research Questions: Resting-state functional MRI (rs-fMRI) offers a non-invasive method to assess brain network integrity. Graph theory–based analyses may enhance prediction of consciousness recovery.
Goals: We evaluated whether combining conventional functional connectivity with graph-theory metrics could predict consciousness recovery via a staged, interpretable statistical framework.
Methods: Rs-fMRI data acquired from 24 patients admitted to an ICU with a DoC, within 1 week post-injury, were analyzed. The primary endpoint was consciousness recovery (command-following) within 6 months. Conventional functional connectivity was measured by mean within-network connectivity (Gordon atlas) with three cut-off strategies: Youden Index, median, and literature-based. We then built a multivariable logistic regression incorporating graph-theory metrics — average path length, network segregation, local clustering coefficient, hub disruption index. The best subset was selected using 5-fold stratified cross-validation.
Results: In patients, the Youden-based threshold of functional connectivity yielded AUC 0.82 with 75% sensitivity and 83% specificity. Adding graph-theoretical metrics improved performance: AUC 0.93, sensitivity 75%, specificity 92%, and precision 90%. Threshold optimization based on Youden Index balanced sensitivity and specificity, producing a generalizable model with approximately 90% accuracy and clinical interpretability.
Conclusion: Combining global connectivity and graph-theory metrics enables sensitive and specific prediction of consciousness recovery. This method provides a foundation for neuroprognostication in hypoxic-ischemic brain injury post-cardiac arrest.