Self-supervised autoencoder network for robust heart rate extraction from noisy photoplethysmogram: Applying blind...

Area of Science:

• Biomedical Engineering
• Signal Processing
• Artificial Intelligence

Background:

• Biosignals are complex mixtures of physiological events.
• Blind Source Separation (BSS) is crucial for extracting individual signals.
• Photoplethysmogram (PPG) signals are often noisy, challenging heart rate (HR) detection.

Purpose of the Study:

• To develop a self-supervised method for separating heartbeat signals from PPG.
• To enhance the accuracy of heart rate detection in noisy PPG data.
• To apply the method to diverse datasets without pre-processing.

Main Methods:

• A self-supervised Multi-Encoder Autoencoder (MEAE) was proposed for BSS.
• The MEAE was trained on a large, open polysomnography database.
• The trained network was applied to noisy PPG datasets from daily activities and surgical patients.

Main Results:

• The separated heartbeat-related source signal significantly improved HR detection.
• The MEAE demonstrated effectiveness on both noisy daily activity and large surgical datasets.
• No pre-processing or data selection was required for training or application.

Conclusions:

• The proposed MEAE method effectively separates heartbeat signals from PPG.
• This approach enhances HR detection accuracy in challenging biosignal conditions.
• The self-supervised, pre-processing-free MEAE shows significant potential for biosignal BSS.