A Machine Learning Driven Pipeline for Automated Photoplethysmogram Signal Artifact Detection

Area of Science:

• Biomedical Engineering
• Signal Processing
• Machine Learning in Healthcare

Background:

• Critical Congenital Heart Disease (CCHD) screening is vital for infant health.
• Photoplethysmography (PPG) signals offer a promising alternative for CCHD detection.
• PPG signal artifacts hinder accurate diagnostic measurements.

Purpose of the Study:

• To develop and evaluate an optimal method for detecting and removing artifacts from PPG waveforms.
• To enhance the reliability of PPG-based CCHD screening.
• To compare artifact detection performance against existing state-of-the-art methods.

Main Methods:

• Feature engineering was applied to PPG signal data.
• Machine Learning (ML) and rule-based algorithms were investigated for artifact detection.
• A 3-stage ML model combining Gradient Boosting (GB) and Random Forest (RF) was proposed.

Main Results:

• The proposed 3-stage ML artifact detection system achieved 84.01% Intersection over Union (IoU).
• Performance is competitive with current state-of-the-art artifact detection techniques.
• The system effectively identifies artifact segments in PPG waveforms.

Conclusions:

• The developed ML-based system offers a robust solution for PPG artifact removal.
• This advancement can improve the accuracy and feasibility of IoT-based CCHD screening.
• Further research can integrate this system into real-time CCHD diagnostic tools.