Abstract
ECG is essential for non-invasive cardiac assessment, yet obtaining standard 12-lead recordings remains challenging in pediatric care because of limited patient cooperation, small chest size, and difficulty with electrode placement. Conventional three-lead gel-electrode ECG is commonly used as a reference method for continuous heart-rate monitoring in newborns and neonatal resuscitation. We propose PedDiffusion, a two-stage generative framework that reconstructs a 12-lead pediatric ECG from single-lead input in maximal-overlap discrete wavelet packet transform (MODWPT) coefficient space. Stage 1 trains a scale-wise U- Net autoencoder to encode a reference lead into multi-band conditioning features. Stage 2 applies a conditional denoising diffusion probabilistic model (DDPM) that jointly generates all 12 leads, guided by frozen reference embeddings across the scale and time dimensions. Inter-lead dependencies are modeled by 2D-CNN layers over the lead-time channels, while cross-scale interaction is confined to attention blocks. Evaluated on the ZZU pECG dataset, our method achieves high reconstruction fidelity (MAE 0.315, RMSE 0.32, PCC 0.924).
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/DT4H_037.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
Open Review Page: https://openreview.net/forum?id=6tAHgq08YC
BibTex
@InProceedings{NguTan_PedDiffusion_MICCAISAT2026,
author = { Nguyen, Tan Anh AND Jun, Hyungmin},
title = { { PedDiffusion: Multi-scale wavelet packet conditional diffusion for pediatric 12-lead ECG reconstruction from a single lead } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17275},
month = {pending},
page = {pending}
}