PECS Framework Enhances Concept Drift Detection in Multimodal Physiologic Signals
A new framework called PECS (Physiologic Stability) has been proposed to detect concept drift in multimodal physiologic signals, particularly focusing on electrocardiogram (ECG) data. The framework compares internal model changes with measurable signal changes to decide whether to keep, change, or flag uncertainty in predictions. It treats ECG as the primary cardiac signal, uses photoplethysmography (PPG) for pulse and vascular information, and incorporates respiration only when ECG and PPG disagree. The framework was tested on the PTB-XL dataset at pilot and full scales, as well as on synchronized BIDMC and MIMIC waveform cohorts. Results showed that the optimal domain pairs varied across datasets, indicating that adding all available signals is not always beneficial. PECS outperformed the evaluated baselines. The paper is available on arXiv with identifier 2608.07759.
Key facts
- PECS is a physiologic stability framework for detecting concept drift in multimodal physiologic signals.
- It compares changes inside the model with measurable changes in the signal.
- ECG is the main cardiac signal; PPG adds pulse and vascular information; respiration is used only when ECG and PPG disagree.
- Tested on PTB-XL (pilot and full scales) and synchronized BIDMC and MIMIC waveform cohorts.
- Optimal domain pairs varied across datasets, showing that adding every available signal is not always best.
- PECS outperformed the evaluated baselines.
- Paper available on arXiv with identifier 2608.07759.
Entities
Institutions
- arXiv