Deep CNN-based detection of cardiac rhythm disorders using PPG signals from wearable devices

Area of Science:

• Biomedical Engineering
• Cardiology
• Artificial Intelligence

Background:

• Cardiac rhythm disorders, including tachycardia, bradycardia, and arrhythmias like atrial fibrillation (AF), pose significant health risks, potentially leading to stroke or sudden death.
• Traditional diagnosis relies on electrocardiography (ECG) and Holter monitors, but wearable devices using photoplethysmography (PPG) offer continuous, accessible monitoring.
• Wearable technology facilitates early detection and intervention for arrhythmias, improving patient outcomes.

Purpose of the Study:

• To develop and evaluate a 1D-Convolutional Neural Network (CNN) model for detecting cardiac arrhythmias using PPG signals.
• To assess the model's accuracy in classifying normal sinus rhythm (NSR), atrial fibrillation (AF), and premature atrial contractions (PAC) against ECG reference data.

Main Methods:

• Utilized a UMMC-prepared dataset containing synchronized ECG and PPG signals.
• Preprocessed ECG signals with a bandpass filter and segmented raw PPG signals into 30-second intervals.
• Trained and validated a 1D-CNN model to classify PPG signals, using ECG as the gold standard.

Main Results:

• The 1D-CNN model achieved a high accuracy of 95.17% in distinguishing between NSR, AF, and PAC using PPG signals.
• Demonstrated the efficacy of PPG signals, when processed by a 1D-CNN, in accurately identifying common cardiac rhythm disorders.
• ECG signals served as a reliable reference for validating the PPG-based detection model.

Conclusions:

• The proposed 1D-CNN model shows significant promise for non-invasive, continuous arrhythmia detection via wearable PPG sensors.
• This approach can empower patients with real-time health monitoring and enable timely medical interventions.
• Further research can explore broader arrhythmia detection and integration into clinical practice.