Boosted-SpringDTW for Comprehensive Feature Extraction of PPG Signals

Area of Science:

• Biomedical Engineering
• Signal Processing
• Machine Learning

Background:

• Accurate feature extraction from physiological signals is crucial for precise parameter estimation.
• Evolving waveform morphologies in signals like photoplethysmography (PPG) pose challenges for traditional methods.
• Wearable devices require robust algorithms for continuous health monitoring.

Purpose of the Study:

• To develop a robust feature extraction framework for physiological signals.
• To enable precise estimation of physiological parameters despite dynamic waveform changes.
• To improve the accuracy of cardiac event identification and interval estimation.

Main Methods:

• Proposed Boosted-SpringDTW, a probabilistic framework combining dynamic time warping (DTW) and heuristics.
• Implemented an automated dynamic template to adapt to evolving waveform morphologies.
• Validated the method on a benchmark PPG dataset with subject- and respiratory-induced variations.

Main Results:

• Achieved precision, recall, and F1-scores over 0.96 for fiducial point identification.
• Obtained mean absolute error less than 11.41 milliseconds for Interbeat Interval (IBI) estimation.
• Demonstrated significant improvements over baseline algorithms in both fiducial point identification and IBI estimation.

Conclusions:

• Boosted-SpringDTW offers superior performance in feature extraction from physiological signals.
• The framework effectively handles evolving waveform morphologies, crucial for wearable applications.
• Enhanced accuracy in physiological parameter estimation supports advanced continuous health monitoring.