On Time Domain Analysis of Photoplethysmogram Signals for Monitoring Heat Stress

Mohamed Elgendi, Rich Fletcher, Ian Norton, Matt Brearley, Derek Abbott, Nigel H. Lovell, Dale Schuurmans2015 · DOI

TL;DR

This paper explores using photoplethysmogram (PPG) signals, specifically the energy of the "aa" area in its second derivative (APG), as a non-invasive method to monitor heat stress in emergency responders. The study found that this feature achieved a promising 79% accuracy in classifying heat-stressed individuals, suggesting a potential for wearable sensors to assess heat stress without traditional, more invasive methods.

What Problem Does This Paper Address?

Heat stress is a significant health concern, particularly for individuals working in hot environments, such as emergency responders. Current methods for assessing heat stress, like measuring body core temperature (BCT) or analyzing heart rate variability (HRV), have limitations. BCT measurements are invasive, and HRV analysis requires lengthy electrocardiogram (ECG) recordings. This creates a need for a simpler, non-invasive, and practical method to monitor heat stress, especially in ambulatory settings. The paper aims to identify the most effective features from PPG signals for this purpose, moving beyond simple heart rate monitoring.

Methods

The study utilizes photoplethysmogram (PPG) signals, which are collected non-invasively from the fingertip. To extract more detailed information, the researchers compute the second derivative of the PPG signal, known as the acceleration photoplethysmogram (APG). They then focus on analyzing time-domain features derived from both PPG and APG waveforms. A key novel aspect is the investigation of the "energy of the aa area" from the APG signal, which represents a specific segment of the waveform. The researchers employed leave-one-out cross-validation and tested four classifiers: Mahalanobis distance, linear discriminant analysis (LDA), quadratic discriminant analysis (QDA), and support vector machine (SVM) to evaluate the performance of different features.

Datasets

The study used PPG data collected from 40 healthy, heat-acclimatized emergency responders (30 males, 10 females) during a simulated chemical, biological, and radiological (CBR) incident in tropical environmental conditions. The data was collected during 30-minute rest periods between exercise sessions. This dataset appears to be specific to this study and not a widely available benchmark.

Key Insights and Ideas

  • PPG as a Heat Stress Indicator: The research demonstrates that PPG signals, beyond just heart rate, contain valuable information for assessing heat stress.
  • APG Waveform Significance: The second derivative of the PPG (APG) provides a more sensitive measure of subtle changes in the waveform related to physiological stress.
  • "Energy of the aa Area" as a Promising Feature: The energy derived from the "aa" segment of the APG waveform emerged as the most informative feature for heat stress detection. This suggests that specific morphological changes in the PPG signal during heat stress are captured by this feature.
  • Non-Invasive Monitoring Potential: The findings highlight the potential for developing wearable PPG-based devices for real-time heat stress monitoring, offering a practical alternative to current methods.
  • Combination of Features: Combining the "energy of the aa area" with traditional HRV metrics (like RMSSD) further improved heat stress detection accuracy, suggesting a synergistic approach.

Key Results

  • The "energy of the aa area" from PPG signals achieved an overall accuracy of 79% in classifying heat-stressed subjects using leave-one-out cross-validation.
  • When combined with RMSSD, the classification accuracy improved to 83%.
  • The study found that the "energy of the aa area" outperformed RMSSD in heat stress detection, with sensitivity and positive predictivity remaining consistent for RMSSD across different exercise stages, while the "energy of the aa area" showed improvement.
  • For the optimal feature (energy of aa), Linear Discriminant Analysis (LDA) showed progressively better performance across exercise stages, reaching an SE of 87.5% and PP of 79.55% after exercise 3.

Limitations and Future Work

  • Generalizability: The study acknowledges that findings need to be explored across different genders, ages, and health conditions to improve generalization.
  • Data Collection Timing: PPG data was collected during rest periods between exercises. Some participants may have cooled down while waiting for measurements, potentially affecting the direct post-exercise stress assessment. Future studies should aim to collect data immediately after exercise.
  • Database Availability: The authors note the lack of publicly available PPG databases measured in tropical conditions or after heat stress, which limits thorough assessment and comparison of algorithms.
  • Future Work: The authors suggest developing a larger study to validate the "energy of the aa area" against traditional heat stress indices and embedding the developed classifier into wearable PPG sensors for real-time stress detection.

Summary updated

BibTeX
@misc{elgendi2015time,
  author = {Mohamed Elgendi and Rich Fletcher and Ian Norton and Matt Brearley and Derek Abbott and Nigel H. Lovell and Dale Schuurmans},
  title = {{On Time Domain Analysis of Photoplethysmogram Signals for Monitoring Heat Stress}},
  year = {2015},
  doi = {10.3390/s151024716},
}
Figure