Intelligent Electrocardiogram Acquisition Via Ubiquitous Photoplethysmography Monitoring

Area of Science:

• Biomedical Engineering
• Machine Learning for Healthcare
• Cardiovascular Monitoring

Background:

• Electrocardiogram (ECG) data are rich for detecting cardiac abnormalities but require active user measurement.
• Photoplethysmography (PPG) is passively collected by consumer devices, enabling continuous monitoring but offers limited diagnostic capability for complex cardiac issues.
• Current PPG analysis is often restricted to basic physiological parameters or obvious arrhythmias like atrial fibrillation.

Purpose of the Study:

• To develop a method that leverages ubiquitous PPG data to identify periods indicative of potential cardiac risk.
• To prompt users for ECG measurements only when PPG data suggests a need for more detailed cardiac assessment.
• To combine the continuous monitoring of PPG with the diagnostic power of ECG for enhanced cardiovascular risk prediction.

Main Methods:

• Development of a dual-convolutional-attention network (DCA-Net) for classifying cardiac abnormalities from PPG signals.
• Training and validation of the DCA-Net model using the MIMIC Waveform Database.
• Testing the model's performance on an independent dataset to assess generalizability.

Main Results:

• The DCA-Net achieved high performance on the MIMIC dataset, with an Area Under the Receiver Operating Characteristic Curve (AUROC) of 0.9 and an Area Under the Precision-Recall Curve (AUPRC) of 0.7.
• Satisfactory results were obtained on an independent dataset, demonstrating an AUROC of 0.7 and AUPRC of 0.6, despite dataset mismatch.
• The study validates the concept of using PPG to intelligently trigger informative ECG measurements.

Conclusions:

• The proposed DCA-Net model demonstrates the feasibility of using passively collected PPG data to infer potential cardiac abnormalities.
• This approach can optimize the use of ECG measurements by prompting them only when necessary, based on PPG signal analysis.
• The findings suggest a promising new paradigm for continuous, accessible cardiovascular risk monitoring using consumer wearable technology.