A Stability- and Aggregation-Based Method for Heart Rate Estimation Using Photoplethysmographic Signals During...

Area of Science:

• Biomedical Engineering
• Signal Processing
• Wearable Technology

Background:

• Photoplethysmography (PPG) is a cost-effective heart rate monitoring alternative to electrocardiography (ECG).
• Deep learning for PPG analysis requires extensive data and computational resources, limiting real-world application, especially without ground truth.
• Motion artifacts significantly degrade PPG signal quality during physical activity.

Purpose of the Study:

• To develop a computationally efficient, one-size-fits-all method for accurate heart rate estimation from PPG during physical activity.
• To minimize motion artifact effects without relying on complex machine learning or deep learning models.
• To demonstrate the efficacy of signal processing techniques in matching deep learning performance.

Main Methods:

• An aggregation-based signal processing approach was employed for heart rate tracking.
• The method was designed to minimize motion artifact impact on PPG signals.
• Evaluation was conducted on multiple public datasets (PPG-DaLiA, WESAD, IEEE) and a new smart ring dataset (UTOKYO).

Main Results:

• The proposed method demonstrated superior performance compared to a CNN ensemble on the PPG-DaLiA and IEEE_Test datasets.
• Mean absolute error (MAE) was reduced by 1.45 bpm and 5.71 bpm on these datasets, respectively.
• The approach achieved high accuracy without requiring extensive computational resources or dataset-specific tuning.

Conclusions:

• Effective signal processing techniques can provide accurate heart rate estimation from PPG during physical activity.
• The developed method offers a practical and resource-efficient alternative to deep learning for wearable health devices.
• This approach overcomes limitations of deep learning, such as data dependency and computational cost.