High-accuracy heart rate detection using multispectral IPPG technology combined with a deep learning algorithm.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Computer Vision

Background:

• Image Photoplethysmography (IPPG) is a noncontact method for physiological parameter detection, widely used for heart rate (HR) monitoring.
• Traditional IPPG methods face limitations including narrow spectral ranges and poor performance in motion detection.
• There is a need for improved noncontact HR monitoring solutions, especially in dynamic environments.

Purpose of the Study:

• To propose an advanced HR detection method utilizing multi-spectral video and IPPG technology.
• To develop and evaluate a novel end-to-end neural network (IPPGResNet18) for real-time HR estimation from facial multi-spectral videos.
• To demonstrate the efficacy of the proposed method in accurately detecting HR under motion conditions.

Main Methods:

• Implementation of a novel HR detection method combining multi-spectral imaging with IPPG technology.
• Development of an end-to-end neural network, IPPGResNet18, for real-time HR evaluation from facial multi-spectral videos.
• Training and validation of the IPPGResNet18 model on a dedicated multi-spectral video dataset.

Main Results:

• The IPPGResNet18 model achieved promising results with MAE = 2.793, RMSE = 3.695, and SD = 3.707.
• The proposed multi-spectral video-based HR detection method demonstrated high accuracy, particularly under motion states.
• Quantitative analysis showed a high degree of agreement (p = 0.304) between estimated and actual HR values.

Conclusions:

• The proposed multi-spectral video-based IPPG method offers a significant advancement for accurate noncontact HR detection.
• The IPPGResNet18 neural network provides a robust solution for real-time HR monitoring, even during subject movement.
• This technology shows clear superiority over conventional methods for real-time HR monitoring during physical activity.