ImputeECG：基于深度学习从不完整记录重建完整12导联心电图

Jul 16, 2026

Upload by Noah

Complete digital 12-lead electrocardiograms (ECGs) are crucial for AI-enabled cardiovascular assessment, yet many clinical ECGs—especially those digitized from ECG images—are incomplete due to short display formats, missing leads, degraded traces, or corrupted signals. ImputeECG is a mask-conditioned one-dimensional Transformer autoencoder that reconstructs complete 12-lead, 10-second ECGs while preserving all observed samples. Training and evaluation on PTB-XL and CPSC2018 are supported by real-world validation on a 43,633-record Kailuan cohort, improving reconstruction errors and downstream diagnostic and demographic prediction performance.

LanguageEnglish

CategoryHealthcare

FormatPDF

Pages23

Reading Time58 min

Abstract

Keywords

Introduction

arXiv :2607 .05009v 1 [ cs .LG] 6 Jul 2026 ImputeECG: Deep Learning Reconstruction of Complete 12-Lead Electrocardiograms from 1 Incomplete Recordings for Cardiac Assessment 2 Xiaocheng Fang 1,2,5 , Haoyu Wang 1,4,5 , Jieyi Cai4 , Qinghao Zhao8 , Jun Li 1,5 , Shanwei Zhang 1,3 , 3 Guangkun Nie 1,2,5 , Yujie Xiao 1,5 , Shun Huang 1,5 , Jiarui Jin 1,2,5 , Hongmin Liu9 , Guodong Wang9 , 4 Shuohua Chen9 , Liming Lin9 , Shouling Wu9 , Hongyan Li2,* , and Shenda Hong 1,5,6,7* 5 1 National Institute of Health Data Science, Peking University, Beijing, China 6 2 School of Intelligence Science and Technology, Peking University, Beijing, China 7 3 Department of Computer Science, Tianjin University of Technology, Tianjin, China 8 4 University of the Chinese Academy of Sciences, Beijing, China 9 5 Institute of Medical Technology, Peking University Health Science Center, Beijing, China 10 6 State Key Laboratory of Vascular Homeostasis and Remodeling, NHC Key Laboratory of 11 Cardiovascular Molecular Biology and Regulatory Peptides, Peking University, Beijing, China 12 7 Institute for Artificial Intelligence, Peking University, Beijing, China 13 8 Department of Cardiology, Peking University People’s Hospital, Beijing, China 14 9 Department of Cardiology, Kailuan General Hospital, Tangshan, China 15 *Correspondence: [hongshenda@pku.edu.cn](hongshenda@pku.edu.cn) 16 Abstract 17 Complete digital 12-lead electrocardiograms (ECGs) are essential for AI-enabled cardiovascular 18 assessment, yet many clinical ECG records, particularly those digitized from ECG images, remain 19 incomplete because of short display formats, incomplete waveform digitization, lead loss, or 20 signal corruption. We developed ImputeECG, a mask-conditioned one-dimensional Transformer 21 autoencoder that completes 12-lead, 10-s ECGs while retaining all observed samples. The model 22 was trained on PTB-XL and evaluated on PTB-XL and CPSC2018 under simulated incomplete 23 settings, with additional real-world validation in a 43,633-record Kailuan clinical cohort after ECG 24 image digitization. Metrics were computed over originally missing regions, with analyses of 25 morphology and downstream diagnostic utility. On PTB-XL, ImputeECG reduced missing-region 26 MAE by 41 .7–51.0% and MSE by 54.0–63.7% versus the strongest baseline, with lower errors in 27 R-peak timing, RR interval, QRS duration, QT interval, and P-wave, QRS-complex, and T-wave 28 reconstruction. On CPSC2018, ImputeECG reduced MAE by 49.7–51.9%, supporting external 29 generalization. In downstream multi-label classification, ImputeECG restored performance to 30 92.28% AUROC and 33.88% AUPRC in the most incomplete PTB-XL setting, approaching 31 complete-ECG performance. On CPSC2018, completed ECGs achieved 94.75–95.89% AUROC 32 and 78.83–81.86% AUPRC across settings. In Kailuan, ECG completion improved zero-shot sex 33 prediction AUROC from 82.6% to 85.8% and reduced age prediction MAE from 10.72 to 9.87 34 years after image-based ECG digitization. These findings support ECG completion as a practical 35 strategy for converting incomplete ECG records into AI-ready 12-lead, 10-s digital signals and 36 extending the usable scope of ECG archives for digital cardiac assessment. 37 Keywords 38 12-lead Electrocardiography, Incomplete ECG Records, ECG Signal Completion, Mask-conditioned 39 Transformer, AI-enabled Cardiac Assessment 40 Introduction 41 The standard 12-lead electrocardiogram (ECG) remains one of the most widely used diagnostic 42 tests in cardiovascular medicine. A complete 10-second digital ECG provides synchronized 43 temporal and spatial information across limb and precordial leads, supporting rhythm interpretation, 44 conduction assessment, ischemia detection, chamber abnormality evaluation, and longitudinal 45 disease monitoring 1–4. With the rapid development of artificial intelligence (AI) for ECG analysis, 46 complete digital waveforms hav

FAQs

Why do many clinical 12-lead ECG records become incomplete?

Incomplete ECGs arise from short display formats, missing leads or waveform regions, and signal corruption or degradation during image-based digitization workflows.

What is ImputeECG and what data does it reconstruct?

ImputeECG is a mask-conditioned one-dimensional Transformer autoencoder that completes complete 12-lead, 10-second ECGs while retaining all observed samples.

How does ECG completion benefit downstream AI tasks in the study?

Completed ECGs restore or improve downstream multi-label diagnostic performance and enable tasks such as demographic prediction (e.g., sex and age) using reconstructed signals.

Jul 16, 2026

Upload by Noah