Real-Time Autoregressive Forecast of Cardiac Features for Psychophysiological Applications.

Area of Science:

• Cardiovascular Physiology
• Biomedical Signal Processing
• Machine Learning in Healthcare

Background:

• Accurate forecasting of cardiac phases is vital for applications like stimulus timing in experiments and physiological predictive modeling.
• Existing autoregressive models have not been applied to predict aortic opening and closing timings.

Purpose of the Study:

• To comparatively analyze autoregressive models for forecasting cardiac timings, specifically aortic opening and closing.
• To evaluate model robustness against noise in seismocardiogram (SCG) signals and feature detection outputs.

Main Methods:

• Utilized various Kalman filter-based autoregressive models.
• Input data included R-peak, aortic opening, and closing timings from electrocardiogram (ECG) and SCG.
• Assessed model performance under introduced noise conditions.

Main Results:

• Time-varying and multi-feature autoregressive algorithms demonstrated superior performance.
• Achieved forecast errors below 2 ms for R-peak, 3 ms for aortic opening, and 10 ms for aortic closing.
• Multi-feature models enhanced noise robustness; time-varying models adapted to physiological changes.

Conclusions:

• Autoregressive models, particularly time-varying and multi-feature approaches, are effective for precise cardiac phase forecasting.
• These models offer significant improvements in noise robustness and adaptability.
• The forecasting capability can be extended to diverse short-term physiological predictive systems beyond cardiac applications.