Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

ECGFlowCMR: Pretraining with ECG-Generated Cine CMR Improves Cardiac Disease Classification and Phenotype Prediction

Xiaocheng Fang, Zhengyao Ding, Jieyi Cai, Yujie Xiao, Bo Liu, Jiarui Jin, Haoyu Wang, Guangkun Nie, Shun Huang, Ting Chen, Hongyan Li, Shenda Hong

TL;DR

This work tackles the scarcity of labeled CMR data by conditioning cine CMR generation on inexpensive ECG signals. The authors introduce ECGFlowCMR, combining a Phase-Aware Masked Autoencoder (PA-MAE) for rhythm-aware ECG representations and an Anatomy-Motion Disentangled Flow (AMDF) that uses a 3D-VAE anatomical template and a diffusion-based flow matcher to synthesize temporally coherent CMR sequences. The method addresses cross-modal temporal mismatch and anatomical observability gap, achieving realistic cine CMR generation and enabling scalable pretraining. Across UK Biobank and a proprietary dataset, ECGFlowCMR improves downstream disease classification and phenotype prediction, with quantitative gains on generation metrics and robust external validation, including a near-chance Turing test suggesting high realism.

Abstract

Cardiac Magnetic Resonance (CMR) imaging provides a comprehensive assessment of cardiac structure and function but remains constrained by high acquisition costs and reliance on expert annotations, limiting the availability of large-scale labeled datasets. In contrast, electrocardiograms (ECGs) are inexpensive, widely accessible, and offer a promising modality for conditioning the generative synthesis of cine CMR. To this end, we propose ECGFlowCMR, a novel ECG-to-CMR generative framework that integrates a Phase-Aware Masked Autoencoder (PA-MAE) and an Anatomy-Motion Disentangled Flow (AMDF) to address two fundamental challenges: (1) the cross-modal temporal mismatch between multi-beat ECG recordings and single-cycle CMR sequences, and (2) the anatomical observability gap due to the limited structural information inherent in ECGs. Extensive experiments on the UK Biobank and a proprietary clinical dataset demonstrate that ECGFlowCMR can generate realistic cine CMR sequences from ECG inputs, enabling scalable pretraining and improving performance on downstream cardiac disease classification and phenotype prediction tasks.

ECGFlowCMR: Pretraining with ECG-Generated Cine CMR Improves Cardiac Disease Classification and Phenotype Prediction

TL;DR

This work tackles the scarcity of labeled CMR data by conditioning cine CMR generation on inexpensive ECG signals. The authors introduce ECGFlowCMR, combining a Phase-Aware Masked Autoencoder (PA-MAE) for rhythm-aware ECG representations and an Anatomy-Motion Disentangled Flow (AMDF) that uses a 3D-VAE anatomical template and a diffusion-based flow matcher to synthesize temporally coherent CMR sequences. The method addresses cross-modal temporal mismatch and anatomical observability gap, achieving realistic cine CMR generation and enabling scalable pretraining. Across UK Biobank and a proprietary dataset, ECGFlowCMR improves downstream disease classification and phenotype prediction, with quantitative gains on generation metrics and robust external validation, including a near-chance Turing test suggesting high realism.

Abstract

Cardiac Magnetic Resonance (CMR) imaging provides a comprehensive assessment of cardiac structure and function but remains constrained by high acquisition costs and reliance on expert annotations, limiting the availability of large-scale labeled datasets. In contrast, electrocardiograms (ECGs) are inexpensive, widely accessible, and offer a promising modality for conditioning the generative synthesis of cine CMR. To this end, we propose ECGFlowCMR, a novel ECG-to-CMR generative framework that integrates a Phase-Aware Masked Autoencoder (PA-MAE) and an Anatomy-Motion Disentangled Flow (AMDF) to address two fundamental challenges: (1) the cross-modal temporal mismatch between multi-beat ECG recordings and single-cycle CMR sequences, and (2) the anatomical observability gap due to the limited structural information inherent in ECGs. Extensive experiments on the UK Biobank and a proprietary clinical dataset demonstrate that ECGFlowCMR can generate realistic cine CMR sequences from ECG inputs, enabling scalable pretraining and improving performance on downstream cardiac disease classification and phenotype prediction tasks.
Paper Structure (10 sections, 8 equations, 3 figures, 4 tables)

This paper contains 10 sections, 8 equations, 3 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Phase-Aware Masked Autoencoder
  4. Anatomy-Motion Disentangled Flow
  5. Experiments
  6. Experiments Setting
  7. Comparisons with State-of-The-Arts
  8. Ablation Studies and Parameter Analysis
  9. External Validation and Qualitative Results
  10. Conclusion

Figures (3)

  • Figure 1: Illustration of our proposed ECGFlowCMR, which integrates PA-MAE and AMDF to synthesize realistic cine CMR sequences from 12-lead ECGs.
  • Figure 2: Ablation study, parameter analysis, and turing test on the UKB dataset.
  • Figure 3: Comparison of synthesized CMR frames across various models.