Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Beyond Motion Artifacts: Optimizing PPG Preprocessing for Accurate Pulse Rate Variability Estimation

Yuna Watanabe, Natasha Yamane, Aarti Sathyanarayana, Varun Mishra, Matthew S. Goodwin

TL;DR

This study shows that non-motion noise in PPG signals can degrade beat-detection and derived metrics even when motion is minimal, challenging the reliance on fixed band-pass filters. By exhaustively exploring 525 low/high cutoff combinations and optimizing filter settings at global and per-person/per-task levels using NSGA-II, the authors demonstrate that adaptive, signal-specific preprocessing substantially improves beat-location accuracy and reduces IBI and PRV errors, with gains up to 7.15% in F1 and reductions of 35 ms (IBI) and 145 ms (RMSSD) compared with a fixed baseline. Across two datasets, adaptive filtering consistently outperforms fixed filters, revealing high inter-individual and context-dependent sensitivity to filter cutoffs and highlighting the need for personalized preprocessing in PPG-based mental-health monitoring. The work discusses factors that shape optimal cutoffs, such as heart rate and stress-related waveform changes, and argues for deploying task- and person-aware filtering in real-world settings where ground-truth ECG is unavailable.

Abstract

Wearable physiological monitors are ubiquitous, and photoplethysmography (PPG) is the standard low-cost sensor for measuring cardiac activity. Metrics such as inter-beat interval (IBI) and pulse-rate variability (PRV) -- core markers of stress, anxiety, and other mental-health outcomes -- are routinely extracted from PPG, yet preprocessing remains non-standardized. Prior work has focused on removing motion artifacts; however, our preliminary analysis reveals sizeable beat-detection errors even in low-motion data, implying artifact removal alone may not guarantee accurate IBI and PRV estimation. We therefore investigate how band-pass cutoff frequencies affect beat-detection accuracy and whether optimal settings depend on specific persons and tasks observed. We demonstrate that a fixed filter produces substantial errors, whereas the best cutoffs differ markedly across individuals and contexts. Further, tuning cutoffs per person and task raised beat-location accuracy by up to 7.15% and reduced IBI and PRV errors by as much as 35 ms and 145 ms, respectively, relative to the fixed filter. These findings expose a long-overlooked limitation of fixed band-pass filters and highlight the potential of adaptive, signal-specific preprocessing to improve the accuracy and validity of PPG-based mental-health measures.

Beyond Motion Artifacts: Optimizing PPG Preprocessing for Accurate Pulse Rate Variability Estimation

TL;DR

This study shows that non-motion noise in PPG signals can degrade beat-detection and derived metrics even when motion is minimal, challenging the reliance on fixed band-pass filters. By exhaustively exploring 525 low/high cutoff combinations and optimizing filter settings at global and per-person/per-task levels using NSGA-II, the authors demonstrate that adaptive, signal-specific preprocessing substantially improves beat-location accuracy and reduces IBI and PRV errors, with gains up to 7.15% in F1 and reductions of 35 ms (IBI) and 145 ms (RMSSD) compared with a fixed baseline. Across two datasets, adaptive filtering consistently outperforms fixed filters, revealing high inter-individual and context-dependent sensitivity to filter cutoffs and highlighting the need for personalized preprocessing in PPG-based mental-health monitoring. The work discusses factors that shape optimal cutoffs, such as heart rate and stress-related waveform changes, and argues for deploying task- and person-aware filtering in real-world settings where ground-truth ECG is unavailable.

Abstract

Wearable physiological monitors are ubiquitous, and photoplethysmography (PPG) is the standard low-cost sensor for measuring cardiac activity. Metrics such as inter-beat interval (IBI) and pulse-rate variability (PRV) -- core markers of stress, anxiety, and other mental-health outcomes -- are routinely extracted from PPG, yet preprocessing remains non-standardized. Prior work has focused on removing motion artifacts; however, our preliminary analysis reveals sizeable beat-detection errors even in low-motion data, implying artifact removal alone may not guarantee accurate IBI and PRV estimation. We therefore investigate how band-pass cutoff frequencies affect beat-detection accuracy and whether optimal settings depend on specific persons and tasks observed. We demonstrate that a fixed filter produces substantial errors, whereas the best cutoffs differ markedly across individuals and contexts. Further, tuning cutoffs per person and task raised beat-location accuracy by up to 7.15% and reduced IBI and PRV errors by as much as 35 ms and 145 ms, respectively, relative to the fixed filter. These findings expose a long-overlooked limitation of fixed band-pass filters and highlight the potential of adaptive, signal-specific preprocessing to improve the accuracy and validity of PPG-based mental-health measures.

Paper Structure

This paper contains 16 sections, 1 equation, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Background & Related Work
  3. Preprocessing and filtering strategies
  4. Adaptive denoising techniques
  5. Methods
  6. Datasets
  7. Initial data preprocessing
  8. Metrics for evaluation
  9. Findings and Evaluation
  10. The non-"motion" artifacts
  11. Evaluating different parameters
  12. Discussion
  13. Impact of cutoff frequencies on RMSSD
  14. Factors influencing optimal cutoffs
  15. The need for adaptive filtering
  16. ...and 1 more sections

Figures (3)

  • Figure 1: Correlations between motion artifacts ($\mathrm{AUC}_\mathrm{MIMS}$) and beat- detection accuracy (F1 score, MAE IBI & RMSSD).
  • Figure 2: F1 score and MAE IBI and RMSSD changes across varying (a) low- cut and (b) high- cut frequencies in the bandpass filter for the Stress- free dataset. Each column represents a task, and each line represents a single participant's metric.
  • Figure 3: Distribution of mean IBI and RMSSD derived from ECGs and PPGs preprocessed with three types of filters for (a) the WESAD and (b) the Stress- free datasets. Brackets with asterisks indicate significant pairwise differences.