BioLT: Streaming Biosignal Modeling

BioLT, a streaming decoder-only framework for long-context biosignal modeling, achieves the highest detection AUROC in 15 of 21 dataset-feature settings and highest 20-minute forecasting AUROC in 13.

• Why it matters: Existing ECG, PPG, and EEG pipelines mostly use short, isolated windows and struggle with long, irregular, interrupted streams, limiting clinically relevant long-range modeling.
• What they did: BioLT converts any window-level feature extractor or pretrained biosignal foundation model into a streaming model using a discontinuity-aware encoding with elapsed-time input and a fixed-size compressed memory bank with cumulative summary statistics.
• The result: On CPSC-2021, VitalDB, MC-MED, and CHB-MIT for arrhythmia, sepsis, and seizure detection and forecasting up to 20 minutes ahead, it improves over state-of-the-art snapshot foundation-model embeddings across 21 dataset-feature settings.

AI summary of the abstract · may contain errors · not medical advice · read the paper.