MAGIC: Multi-Grained Conditional Diffusion for High-Fidelity PPG-to-ECG Translation

Area of Science:

• Biomedical Engineering
• Computational Biology
• Artificial Intelligence

Background:

• High-fidelity electrocardiogram (ECG) signals are crucial for accurate cardiac digital twins.
• Conventional ECG is unsuitable for long-term home monitoring.
• Photoplethysmography (PPG) offers continuous data but lacks electrophysiological detail.

Purpose of the Study:

• To propose MAGIC, a multi-grained conditional diffusion framework.
• To generate ECG-like waveforms from PPG signals for computational analysis.
• To bridge the gap between PPG and ECG modalities for improved cardiovascular simulation.

Main Methods:

• MAGIC utilizes a diffusion Transformer with multi-grained conditional information.
• Global conditions summarize cardiovascular context via adaptive layer normalization.
• Local conditions preserve PPG cues through gated cross-attention.
• Conditional encoder pre-trained using ROI-weighted reconstruction, contrastive learning, and latent alignment.

Main Results:

• MAGIC improves distributional fidelity across public datasets and downstream tasks.
• Pre-training significantly reduces Frechet Distance (FD) from 0.908 to 0.693 on MIMIC-AFib.
• MAGIC achieves lower FD with fewer sampling steps compared to RDDM.

Conclusions:

• MAGIC effectively generates ECG-like waveforms from PPG signals.
• The framework enhances the utility of PPG for cardiovascular computational analysis.
• MAGIC facilitates the development of cardiac digital twins for personalized health monitoring.