Low-density EEG-based Functional Connectivity Discriminates Minimally Conscious State plus from minus

Objective: Within the continuum of consciousness, patients in a Minimally Conscious State (MCS) may exhibit high-level behavioral responses (MCS+) or may not (MCS−). The evaluation of residual consciousness and related classification is crucial to propose tailored rehabilitation and pharmacological treatments, considering the inherent differences among groups in diagnosis and prognosis. Currently, differential diagnosis relies on behavioral assessments posing a relevant risk of misdiagnosis. In this context, EEG offers a non-invasive approach to model the brain as a complex network. The search for discriminating features could reveal whether behavioral responses in post-comatose patients have a defined physiological background. Additionally, it is essential to determine whether the standard behavioral assessment for quantifying responsiveness holds physiological significance. Methods: In this prospective observational study, we investigated whether low-density EEG-based graph metrics could discriminate MCS+/− patients by enrolling 57 MCS patients (MCS−: 30; males: 28). At admission to intensive rehabilitation, 30 min resting-state closed-eyes EEG recordings were performed together with consciousness diagnosis following international guidelines. After EEG preprocessing, graphs’ metrics were estimated using different connectivity measures, at multiple connection densities and frequency bands ($\alpha ,\theta ,\delta$). Metrics were also provided to cross-validated Machine Learning (ML) models with outcome MCS+/−. Results: A lower level of brain activity integration was found in the MCS− group in the $\alpha$ band. Instead, in the $\delta$ band MCS− group presented an higher level of clustering (weighted clustering coefficient) respect to MCS+. The best-performing solution in discriminating MCS+/− through the use of ML was an Elastic-Net regularized logistic regression with a cross-validation accuracy of $79\%$ (sensitivity and specificity of $74\%$ and $85\%$ respectively). Conclusion: Despite tackling the MCS+/− differential diagnosis is highly challenging, a daily-routine low-density EEG might allow to differentiate across these differently responsive brain networks. Significance: Graph-theoretical features are shown to discriminate between these two neurophysiologically similar conditions, and may thus support the clinical diagnosis.