Evaluation of EEG Foundation Models for Event-Based Burst-Suppression Detection in ICU

The study evaluates EEG Foundation Models (FMs) for detecting burst suppression (BS) in ICU settings, specifically using reduced-montage EEG without patient-specific calibration. The REVE-base model achieved the highest event-based F1-score of 0.868, significantly outperforming both EEGNet and an adaptive thresholding baseline in burst detection accuracy. This research highlights the potential of FMs for scalable EEG monitoring in clinical environments, particularly in scenarios with limited annotated data, emphasizing the effectiveness of full fine-tuning and pretrained models for improving detection performance.