Key points are not available for this paper at this time.
Background Burst suppression (BS) is a clinically important EEG pattern in disorders of consciousness (DoC), but routine identification still relies on visual review, and automated methods developed in anesthesia or otherwise controlled settings may generalize poorly to heterogeneous DoC EEG. Objective To develop an adaptive framework for automated BS detection and algorithm-aligned BS-burden quantification, and to characterize BS-associated neurophysiological features in DoC. Methods We developed an unsupervised two-stage pipeline that screens EEG segments for BS using differential-signal enhancement and adaptive thresholding, then quantifies BS-positive segments using a burst-to-suppression ratio ( R bs ). Segment-level performance was evaluated on one baseline validation dataset and two stress-test datasets, each including 63 BS and 192 non-BS segments, and benchmarked against two re-implemented detectors. In the DoC-BS cohort, spectral, complexity, and phase-synchrony features were extracted. Patient-level R bs was correlated with same-day and day-7 total Glasgow Coma Scale (GCS) scores; exploratory feature-level correlations with day-7 GCS and R bs were also examined. Results The detector achieved strong performance across datasets (F1, 93.9%–98.4%; MCC, 0.920–0.979) and fewer false positives than the comparator methods. BS was characterized by frontal δ predominance, elevated DAR, reduced SampEn, denser δ and θ phase synchrony, and weaker α-band coupling. R bs showed positive nominal associations with same-day GCS and day-7 GCS, whereas exploratory feature-level correlations did not survive BH-FDR correction. Conclusion This study provides an adaptive BS screening-and-quantification framework for heterogeneous DoC EEG. R bs is an objective descriptor of BS burden, and associated EEG features represent candidate neurophysiological signatures requiring multicenter validation.
Li et al. (Mon,) studied this question.