Heart rate estimation in PPG signals using Convolutional-Recurrent Regressor.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Machine Learning

Background:

• Photoplethysmography (PPG) is a common method for heart rate monitoring.
• Traditional spectral analysis of PPG signals degrades with motion artifacts, limiting accuracy during physical activity.
• Deep learning offers potential for robust signal analysis in challenging conditions.

Purpose of the Study:

• To develop an effective heart rate estimation method from PPG signals for subjects undergoing exercise.
• To address the performance limitations of spectral analysis caused by motion artifacts.
• To leverage deep learning techniques for improved PPG-based heart rate monitoring.

Main Methods:

• Feature extraction from PPG signals considering signal, spectral, and time-frequency properties.
• Implementation of a hybrid convolutional-recurrent neural network (C-RNN) model.
• Training and evaluation of the C-RNN in a regression framework for heart rate estimation.

Main Results:

• The proposed C-RNN method achieved low error rates on the IEEE signal processing cup dataset.
• Subject-dependent protocol error rate was 2.41 ± 2.90 bpm.
• Subject-independent protocol error rate was 3.8 ± 2.3 bpm.

Conclusions:

• The hybrid C-RNN model demonstrates effectiveness in estimating heart rate from PPG signals during exercise.
• The method successfully mitigates the impact of motion artifacts, enhancing monitoring reliability.
• This approach validates the use of deep learning for robust PPG-based heart rate monitoring in dynamic conditions.