A Robust PPG Signal Processing Pipeline for Biomarker Extraction and Analysis

Photoplethysmography (PPG) is a non-invasive optical measurement method widely used for monitoring cardiovascular health. PPG signals provide a rich source of continuoustime data, valuable for various applications including circuit and system design, biomarker extraction, and physiological analysis. The availability of large datasets like the Medical Information Mart for Intensive Care III(MIMIC-III) dataset provides opportunities to develop advanced signal-processing algorithms for PPG data and offers great potential for artificial intelligence (AI) applications in healthcare. Recognizing the inherent variability in PPG signal quality, we present a comprehensive PPG signal processing pipeline designed to provide researchers with a tool for collecting and analyzing high-quality PPG data. The pipeline utilizes the MIMIC-III matched waveform dataset and the pyPPG Python toolbox. We demonstrate a workflow for PPG signal extraction, signal quality assessment using dynamic time wrapping-based (DTW) signal quality indexing (SQI), and fiducial point and biomarker extraction. The proposed pipeline provides researchers and developers with high-quality, labeled PPG segments and extracted features. We processed a total of 1.83 million (2.4 TB) PPG segments and successfully extracted 83,597 high-quality 60 -second segments (16 GB). For each signal segment, we extracted its fiducial points and a total of 102 pulse wave features using the pyPPG toolbox. We analyze the distributions of key biomarkers and discuss the pipeline's effectiveness. This work offers a standardized approach to signal processing that can enhance the reliability and efficiency of PPGbased health assessment tools and lay the foundation for future efforts in developing advanced algorithms for wearable health monitoring systems.