A Novel Machine-Learning-Based Noise Detection Method for Photoplethysmography Signals.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Machine Learning

Background:

• Wearable devices are crucial for continuous health monitoring and early disease detection.
• Physiological signals from wearables are prone to interference like Motion Artifacts (MA) and Baseline Wanders (BW).
• Accurate noise detection in wearable signals is vital to prevent false alarms in remote healthcare.

Purpose of the Study:

• To develop a Machine Learning (ML)-based method for identifying noise in Photoplethysmogram (PPG) signals.
• To distinguish between clean, MA-corrupted, and BW-corrupted PPG signal segments.
• To evaluate the performance of the ML approach without relying on data from other sensors.

Main Methods:

• A Machine Learning (ML) model was trained to classify PPG signal segments.
• The model was designed to differentiate between clean signals and those corrupted by Motion Artifacts (MA) or Baseline Wanders (BW).
• The approach was validated without using auxiliary sensor data, such as accelerometer readings.

Main Results:

• The ML-based classification achieved high accuracy in distinguishing signal types.
• F1-scores ranged from 89.3% for three-class classification to 99.4% for binary classification.
• The proposed method effectively identified noise in PPG signals without external sensor references.

Conclusions:

• The developed ML approach offers a robust solution for detecting artifacts in wearable PPG signals.
• This method can enhance the reliability of remote health monitoring systems by reducing false alarms.
• The sensor-independent nature of this technique makes it broadly applicable to various wearable healthcare devices.