DeepCRI: real-time EEG-based prognostication after cardiac arrest.

Abstract

Accurate prediction of neurological outcome after cardiac arrest is essential for guiding intensive care decisions. Electroencephalography (EEG) supports prognostication; however, interpretation relies on expert judgment and is often subjective and delayed. We developed DeepCRI, a bedside-integrated deep learning system that produces continuously updated prognostic trajectories during the first 36 h after arrest. DeepCRI uses time-dependent decision boundaries to define good-, poor-, and gray-zone regions over time, and applies a lock-in rule that fixes classification only after sustained, concordant high-confidence evidence within a compact temporal window, thereby preventing transient threshold crossings from driving decisions. DeepCRI was developed on a multicenter EEG dataset of 522 comatose patients after cardiac arrest and subsequently evaluated in independent internal (n = 219) and external validation cohorts (n = 167). In the internal validation cohort, DeepCRI provided lock-in classifications in 179/219 patients (81.7%), with a sensitivity of 94.7% (95% CI 90.0-97.6%) and specificity of 81.9% (95% CI 73.5-88.1%) for good outcome, and a sensitivity of 49.5% (95% CI 40.2-58.9%) and specificity of 100.0% (95% CI 96.7-100.0%) for poor outcome; 40/219 patients (18.3%) remained in the gray zone. In the external validation cohort, DeepCRI provided lock-in classifications in 100/167 patients (59.9%), with a sensitivity of 67.2% (95% CI 54.7-77.7%) and specificity of 82.1% (95% CI 73.7-88.2%) for good outcome, and a sensitivity of 36.8% (95% CI 28.2-46.3%) and specificity of 98.4% (95% CI 91.3-99.7%) for poor outcome. Post hoc analysis indicated residual EMG artifacts contributed to this false poor-outcome prediction. By embedding DeepCRI into routine ICU EEG infrastructure, we demonstrate the technical feasibility and clinical promise of continuous, real-time AI-driven prognostication for comatose patients after cardiac arrest.