A Supervised Approach to Robust Photoplethysmography Quality Assessment

Area of Science:

• Biomedical Engineering
• Cardiology
• Signal Processing

Background:

• Early detection of Atrial Fibrillation (AFib) is critical for stroke prevention.
• Photoplethysmogram (PPG) from wearable devices is increasingly used for long-term AFib monitoring.
• Traditional PPG quality assessment methods struggle with arrhythmias like AFib due to pulse morphology changes.

Purpose of the Study:

• To evaluate existing PPG quality assessment algorithms for AFib detection.
• To develop and compare machine learning approaches for robust PPG quality assessment in AFib patients.
• To determine the effectiveness of PPG quality assessment for stroke risk stratification.

Main Methods:

• Tested existing PPG quality assessment algorithms on AFib patient data.
• Developed and compared supervised machine learning models (k-NN, decision trees, SVM) for PPG quality assessment.
• Utilized fingertip PPG (fPPG) and radial PPG (rPPG) data from intensive care unit patients.

Main Results:

• Support Vector Machine (SVM) demonstrated superior performance in PPG quality assessment.
• The SVM model achieved 0.9477 accuracy on fPPG data and 0.9589 accuracy on rPPG data.
• Machine learning models effectively assessed PPG quality despite AFib-induced signal variations.

Conclusions:

• Machine learning, especially SVM, offers a robust solution for PPG signal quality assessment in AFib.
• Improved PPG quality assessment can enhance the reliability of wearable devices for AFib detection and stroke prevention.
• This approach supports better risk stratification and personalized monitoring for patients at risk of stroke.