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

2026-06-18 · Source: Artificial Intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Medical AI Applications · Depth: Expert, quick

Summary

A study evaluated EEG Foundation Models (FMs) for event-based burst-suppression (BS) detection in Intensive Care Unit (ICU) electroencephalography (EEG) using reduced-montage data and no patient-specific calibration. The research compared REVE-base, LUNA-large, and LuMamba-Tiny against an adaptive thresholding baseline and an EEGNet baseline. REVE-base emerged as the top performer, achieving an event-based F1-score of 0.868 \u00b1 0.167. This model also significantly reduced burst-per-minute error by 52.1% compared to EEGNet and 36.2% against adaptive thresholding. Ablation experiments indicated that full fine-tuning was the most effective adaptation strategy, improving LUNA-large's event-based F1-score by up to +0.102 over frozen-backbone training. Furthermore, pretrained REVE-base demonstrated a substantial advantage, outperforming random initialization by +0.723 event-based F1 points when trained with only 25% of the labeled cohort.

Key takeaway

For research scientists developing automated EEG monitoring in critical care, you should prioritize evaluating foundation models like REVE-base for burst-suppression detection. These models offer robust performance, achieving an F1-score of 0.868 \u00b1 0.167 without patient-specific calibration, and significantly reduce error rates. When adapting these models, focus on full fine-tuning, as it proved most effective. Leveraging pretrained models is crucial for maintaining high performance, even when your labeled datasets are limited.

Key insights

EEG Foundation Models show promise for robust, patient-agnostic burst-suppression detection in ICU settings, even with limited data.

Principles

• Full fine-tuning is superior for FM adaptation.
• Pretraining significantly boosts performance with scarce data.

Method

The study evaluated FMs for event-based burst detection in reduced-montage ICU EEG, comparing full fine-tuning, frozen-backbone, two-step, and LoRA adaptation strategies.

In practice

• Consider REVE-base for ICU EEG monitoring.
• Prioritize full fine-tuning for FM deployment.
• Leverage pretraining with limited labeled data.

Topics

• EEG Foundation Models
• Burst Suppression Detection
• Intensive Care Units
• REVE-base
• Model Fine-tuning
• Pretraining

Best for: AI Scientist, Research Scientist

Related on AIssential

• Deep Sleep Classification via EEG Signal Criticality: A Passive BCI Approach for Sleep-Improvement Neurofeedback — High-accuracy passive BCI sensing is achievable via probabilistic decoding of EEG criticality.
• CogAdapt: Transferring Clinical ECG Foundation Models to Wearable Cognitive Load Assessment via Lead Adaptation — Adapting clinical ECG foundation models via lead transformation and progressive fine-tuning enables robust wearable cognitive load assessment.
• Neural Response to Familiar Names Predicts Outcome of Comatose ICU Patients: A Prospective Observational Cohort Study — EEG responses to familiar names predict outcomes in comatose ICU patients.
• Toward Training-Free Zero-Shot Anomaly Detection in 3D Medical Images: A Batch-Based Approach Using 2D Foundation Models — CS3F enables training-free zero-shot anomaly detection in 3D medical images by adapting 2D foundation models.
• Beyond Augmentation: Score-Guided Pathological Prior for EEG-based Depression Detection — A score-guided classification framework improves EEG-based depression detection by modeling pathological priors without data augmentation.
• Active Evidence-Seeking and Diagnostic Reasoning in Large Language Models for Clinical Decision Support — LLMs exhibit a substantial performance gap in active, multi-turn clinical diagnosis compared to static, full-context evaluations.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.