Robust PPG motion artifact detection using a 1-D convolution neural network.

Area of Science:

• Biomedical Engineering
• Signal Processing
• Machine Learning

Background:

• Photoplethysmography (PPG) is crucial for continuous physiological monitoring via wearable sensors.
• PPG signals are prone to artifacts, compromising the accuracy of derived parameters like heart rate and oxygen saturation.

Purpose of the Study:

• To develop a deep learning model for automated classification of PPG segments into clean or artifact-affected.
• To avoid manual feature engineering and data-dependent pulse segmentation in artifact detection.

Main Methods:

• A 13-layer one-dimensional Convolutional Neural Network (1-D-CNN) was designed to automatically learn PPG waveform features.
• The model was trained and validated on a local dataset and evaluated on two independent PhysioNet MIMIC II datasets.
• Blind signal processing and Z-score normalization were employed for segment preparation.

Main Results:

• The 1-D-CNN model achieved high testing accuracy, reaching 94.9% on the local dataset and 93.8%, 86.7% on independent datasets.
• The model demonstrated strong generalization capabilities across multiple cohorts, with an overall accuracy of 94.5%.
• Comparable performance was achieved against existing reported methods.

Conclusions:

• Deep learning and blind signal processing effectively detect motion artifacts in PPG signals.
• The proposed 1-D-CNN method avoids manual feature engineering, offering a robust and generalizable solution for PPG artifact detection.