PPG-to-ECG Signal Translation for Continuous Atrial Fibrillation Detection via Attention-based Deep State-Space...

Area of Science:

• Biomedical Engineering
• Signal Processing
• Artificial Intelligence in Healthcare

Background:

• Photoplethysmography (PPG) is a non-invasive optical technique for measuring cardiac physiology, widely used in wearable devices.
• Despite its popularity, PPG lacks the substantial clinical diagnostic value of electrocardiography (ECG).
• A strong correlation exists between PPG and ECG signals, suggesting potential for PPG-based ECG reconstruction.

Purpose of the Study:

• To develop a subject-independent model for translating PPG signals into corresponding ECG waveforms.
• To create a robust and data-efficient model capable of handling noisy real-world data.
• To enable the integration of PPG's continuous monitoring with ECG's diagnostic capabilities for early cardiovascular disease detection.

Main Methods:

• Proposed a subject-independent attention-based deep state-space model (ADSSM).
• Incorporated probabilistic prior knowledge for data efficiency and noise robustness.
• Evaluated the model using 55 subjects' data from the MIMIC-III database, including noisy variations.

Main Results:

• The ADSSM demonstrated effectiveness in translating PPG to ECG signals.
• Translated ECG signals achieved a PR-AUC of 0.986 when used with an existing atrial fibrillation (AFib) detector.
• The model proved robust to noise and efficient with limited data.

Conclusions:

• The ADSSM successfully translates PPG signals to ECG waveforms, offering a new avenue for cardiovascular monitoring.
• This approach bridges the gap between PPG's accessibility and ECG's diagnostic power.
• Enables early diagnosis of cardiovascular diseases by combining continuous PPG measurements with ECG's established knowledge base.