Robust PPG Peak Detection Using Dilated Convolutional Neural Networks.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Machine Learning

Background:

• Accurate photoplethysmogram (PPG) signal peak determination is crucial for physiological monitoring, including heart rate calculation.
• Conventional peak detection methods struggle with low signal-to-noise ratio (SNR) PPG data, often corrupted by noise and motion artifacts.
• Existing algorithms are insufficient for robustly analyzing PPG signals in real-world, free-living conditions.

Purpose of the Study:

• To enhance the noise-resiliency of PPG signal analysis.
• To propose and validate a robust peak detection algorithm for noisy PPG signals, including those with motion artifacts.
• To improve the accuracy of physiological parameter extraction from low-SNR PPG data.

Main Methods:

• Development of a novel peak detection algorithm utilizing convolutional neural networks (CNNs) with dilated convolutions.
• Training and evaluation using a smartwatch-collected PPG dataset from a home-based health monitoring application.
• Implementation of a data generator to create synthetic noisy PPG data for comprehensive model assessment across various SNRs (0-45 dB).

Main Results:

• The proposed CNN-based algorithm significantly outperforms conventional adaptive threshold, transform-based, and existing machine learning methods.
• Achieved overall precision, recall, and F1-score of 82%, 80%, and 81%, respectively, across all tested SNR ranges.
• Demonstrated superior performance compared to state-of-the-art methods, which yielded a best of 78% precision, 80% recall, and 79% F1-score.

Conclusions:

• The developed algorithm provides accurate PPG peak detection, even in the presence of substantial noise and motion artifacts.
• This robust method enhances the reliability of PPG signal analysis for home-based health monitoring applications.
• The findings support the use of advanced deep learning techniques for improving physiological signal processing in challenging environments.