Chain of Flow: A Foundational Generative Framework for ECG-to-4D Cardiac Digital Twins

TL;DR

Chain of Flow is introduced, a foundational ECG-driven generative framework that reconstructs full 4D cardiac structure and motion from a single cardiac cycle and transforms cardiac digital twins from narrow predictive models into fully generative, patient-specific virtual hearts.

Abstract

A clinically actionable Cardiac Digital Twin (CDT) should reconstruct individualised cardiac anatomy and physiology, update its internal state from multimodal signals, and enable a broad range of downstream simulations beyond isolated tasks. However, existing CDT frameworks remain limited to task-specific predictors rather than building a patient-specific, manipulable virtual heart. In this work, we introduce Chain of Flow (COF), a foundational ECG-driven generative framework that reconstructs full 4D cardiac structure and motion from a single cardiac cycle. The method integrates cine-CMR and 12-lead ECG during training to learn a unified representation of cardiac geometry, electrophysiology, and motion dynamics. We evaluate Chain of Flow on diverse cohorts and demonstrate accurate recovery of cardiac anatomy, chamber-wise function, and dynamic motion patterns. The reconstructed 4D hearts further support downstream CDT tasks such as volumetry, regional function analysis, and virtual cine synthesis. By enabling full 4D organ reconstruction directly from ECG, COF transforms cardiac digital twins from narrow predictive models into fully generative, patient-specific virtual hearts. Code will be released after review.

Figures (8)

• Figure 1: Overview of the COF for ECG-driven cardiac digital twin generation. The framework integrates CMR volumes and 12-lead ECG during training to learn a unified representation of cardiac anatomy, electrophysiology, and motion dynamics. Cine-CMR provides anatomical geometry and motion supervision, while ECG supplies subject-specific electrophysiological dynamics over a single cardiac cycle. These multimodal signals are fused within the proposed Chain of Flow, which couples geometry encoding, electro-physiological flow, and motion propagation to form a unified ECG-driven generative representation. At inference, the learned model reconstructs a patient-specific 4D cardiac digital twin directly from ECG, yielding individualised cardiac anatomy and motion. The resulting digital heart supports downstream CDT applications, including chamber volumetry, regional functional analysis, and virtual cine simulation.
• Figure 2: Algorithmic pipeline of physiological motion modelling and ECG-conditioned digital heart construction. The pipeline starts with a topology-preserving volumetric registration module that estimates 3D displacement fields between source and target CMR volumes at different cardiac phases. Discrete deformation samples obtained from registration are parameterised as a continuous spatiotemporal velocity field. A segmentation-guided teacher supervision hub constrains the registration using anatomy-aware Dice loss. The resulting subject-specific motion dynamics are used as supervision for flow matching, linking ECG signals to physiological motion priors. Finally, the learned velocity field is integrated over time via ODE-based integration to produce a 4D digital heart with patient-specific cardiac motion.
• Figure 3: ECG-conditioned digital heart generation results. A single R-R interval extracted from the 12-lead electrocardiogram is used as the temporal conditioning signal (top), where the R, S, T, and P waves indicate distinct cardiac phases. The R wave corresponds to the ED phase, while the T wave corresponds to the ES phase. On the left, the anatomical slice stack context is illustrated, showing the volumetric coverage of short-axis multi-slice CMR acquired across the heart. On the right, ECG-conditioned multi-slice CMR results are shown. For each selected cardiac phase, four rows are displayed: the first row shows the GT multi-slice CMR images, and the second row shows the corresponding ECG-conditioned generated slices. The third row presents the GT segmentation masks, while the fourth row shows the segmentation results obtained from the generated CMR slices.
• Figure 4: (a) Radar-based comprehensive comparison across six image-level metrics (SSIM, PSNR, FID, FVD, M-Corr., and M-SSIM), where all indicators are normalised so that larger area indicates better overall performance. (b) Segmentation-based comparison using per-frame Dice and IoU for the left ventricle, right ventricle, and myocardium.
• Figure 5: Per-slice segmentation accuracy across cardiac phases. Per-slice Dice, HD95, and IoU are reported at the end-diastolic and end-systolic phases. Slices are ordered by slice rank from basal to apical (slice ranks 2–7). For each phase, metrics are computed separately for the left ventricle, right ventricle, and myocardium. Dice and IoU heatmaps (a, b, e, f) use a normalised scale from 0 to 1, while HD95 heatmaps (c, d) report 2D per-slice surface distances computed using in-plane spacing. Each cell represents the mean value across all subjects for the corresponding anatomical structure and slice rank.