Binary Classification of Consciousness Using Cerebral Blood Flow and EEG Features

Assessing consciousness at the bedside in the neurocritical care unit is complicated by sedation and other treatment effects. While electroencephalogram (EEG) is commonly used, it offers a limited view of the neurovascular unit. We evaluated whether combining cerebral blood flow (CBF) features with EEG improves binary classification of consciousness in patients with severe brain injury.

We retrospectively analyzed 26 adults who underwent multimodal neuromonitoring. Signals were segmented into 30-min windows after each probe recalibration. We used parameters including CBF low-frequency bands (band IV 0.027-0.073 Hz, band V 0.01-0.027 Hz, and band all 0-0.5 Hz) and EEG band powers (delta-beta), alpha-delta ratio (ADR), alpha/(delta + theta) (ADTR), total power. A random forest (RF) model trained using K-fold cross-validation achieved optimal classification. Highly correlated features (r > 0.8) were excluded from simultaneous use. Performance was summarized with receiver operating characteristic-area under the curve (ROC-AUC) and accuracy, with confusion matrices shown for the best combinations.

Multimodal feature combinations significantly improved classification compared with EEG features alone. The best-performing combination (EEG ADR, total EEG power, and CBF B and V) achieved a ROC-AUC of 0.86 and an accuracy of 82%, outperforming EEG-only models. In two exploratory cases with noninvasive diffuse correlation spectroscopy (DCS) monitoring, the same feature framework produced predictions concordant with those obtained using invasive CBF features.

Combining EEG and CBF metrics, particularly low-frequency oscillations in perfusion fluctuations, enhances classification of consciousness in critically ill patients and may support future bedside tools for real-time neurovascular monitoring and decision making about treatment and rehabilitation.