Physics-Informed 12-Lead ECG Reconstruction
Physics-Informed Conditional Diffusion Framework
This study develops a conditional diffusion framework that reconstructs the full 12-lead ECG from a single wearable lead. Cross-attention conditioning and an Aliev–Panfilov cardiac-potential prior guide generation toward physiologically consistent waveforms.
Benchmark Accuracy and Waveform Fidelity
On PTB-XL, the model achieves the lowest error among reported baselines, with MAE 0.1153 and MSE 0.0546. Reconstructions recover STEMI segments, preserve R-peak magnitude, and remain robust without baseline-wander filtering.
