End-to-End 4D Heart Mesh Recovery Across Full-Stack and Sparse Cardiac MRI

Jul 16, 2026

Upload by Himbo

Reconstructing cardiac motion from CMR sequences underpins diagnosis, prognosis, and image-guided intervention. Many existing approaches require complete CMR stacks, making them impractical during procedures when only sparse, low-resolution slices are available. TetHeart provides a unified end-to-end framework for 4D heart mesh recovery from both offline full-stack and intra-procedural sparse-slice observations. It uses deformable tetrahedra, slice-adaptive 2D–3D feature assembly, sparsity-focused distillation, and weakly supervised keyframe motion learning, achieving state-of-the-art accuracy without retraining.

LanguageEnglish

CategoryHealthcare

FormatPDF

Pages26

Reading Time1 h 6 min

Introduction

Motivation for intervention-time reconstruction

TetHeart unified 4D mesh recovery framework

arXiv :2509 . 12090v3 [ cs .CV] 12 Jul 2026 End-to-End 4D Heart Mesh Recovery Across Full-Stack and Sparse Cardiac MRI Yihong Chen yihong. chen@epfl. ch School of Computer and Communication Science, EPFL [Jiancheng Yang](Jiancheng Yang jiancheng.yang@aalto.fi)[ jiancheng.yang@aalto.fi](Jiancheng Yang jiancheng.yang@aalto.fi) ELLIS Institute Finland, Aalto University, Finland [Deniz Sayin Mercadier](Deniz Sayin Mercadier deniz. mercadier@epfl.ch@epfl. ch)[ deniz. mercadier@epfl.ch@epfl. ch](Deniz Sayin Mercadier deniz. mercadier@epfl.ch@epfl. ch) School of Computer and Communication Science, EPFL Hieu Le [hle40@charlotte. edu](hle40@charlotte. edu) University of North Carolina at Charlotte Juerg Schwitter jurg. schwitter@chuv. ch Center for Interventional MRI, CHUV [Pascal Fua](Pascal Fua pascal.fua@epfl. ch)[ pascal.fua@epfl. ch](Pascal Fua pascal.fua@epfl. ch) School of Computer and Communication Science, EPFL Reviewed on OpenReview: [https: // openreview. net/ forum? id= 9k00kN5yk2](https: // openreview. net/ forum? id= 9k00kN5yk2) Abstract Reconstructing cardiac motion from CMR sequences is critical for diagnosis, prognosis, and intervention. Existing methods rely on complete CMR stacks to infer full heart motion, limiting their applicability during intervention when only sparse observations are available. We present TetHeart, the first end-to-end framework for unified 4D heart mesh recovery from both offline full-stack and intra-procedural sparse-slice observations. Our method leverages deformable tetrahedra to capture shape and motion in a coherent space shared across cardiac structures. Before a procedure, it initializes detailed, patient-specific heart meshes from high-quality full stacks, which can then be updated using whatever slices can be obtained in real time, down to a single slice during the procedure. TetHeart incorporates several key innovations: (i) an attentive slice-adaptive 2D–3D feature assembly mechanism that integrates information from arbitrary numbers of slices at any position; (ii) a distillation strategy to ensure accurate reconstruction under extreme sparsity; and (iii) a weakly supervised motion learning scheme requiring annotations only at keyframes, such as the end-diastolic and end-systolic phases. Trained and validated on three large public datasets, and evaluated on additional private interventional and public datasets without retraining, TetHeart achieves state-of-the-art accuracy in both pre- and intra-procedural settings. Code and dataset is available at [https://github.com/Scalsol/TetHeart](https://github.com/Scalsol/TetHeart). 1 Introduction Modeling cardiac shape and motion is a vital 4D reconstruction problem with direct implications for medical imaging, motion analysis, and image-guided intervention Sanz & Fayad (2008); Li et al. (2024); Qiao et al.(2025); Qian et al. (2025) . Cardiac magnetic resonance (CMR) provides high-quality temporal anatomy and has been widely used for both shape reconstruction Bai et al. (2015); Ye et al. (2023); Dou et al. (2024); Yang et al. (2024); Xiao et al. (2024) and motion estimation Qin et al. (2018); Meng et al. (2022a;b; 2023); Yuan et al. (2023) . However, as shown in Fig. 1 (a), existing methods rely on full volumetric CMR stacks and Previous: Standard Offline Scenario assume availability of full CMR stacks, which can only be obtained offline. Ours: In the Intervention Room only a handful (e.g. , 1-3) of real-time slices can be captured. (a) (b) (c) Figure 1: Cardiac motion reconstruction. (a) Standard offline scenario. (b) Intervention room setting. (c) Interventional MRI setup at our collaborating hospital: A. Preparation room located next to the iCMR room. B. iCMR room equipped with screens to visualize the heart and catheters during the intervention. C. Preparation of the intubated and ventilated patient inside the iCMR. D. Control room with visualizations. are thus restricted to offline analys

FAQs

Why do existing 4D cardiac motion methods struggle during interventions?

They typically assume availability of full volumetric CMR stacks, but intervention-time imaging often provides only a few real-time, sparse slices.

What is TetHeart designed to do in iCMR scenarios?

TetHeart enables consistent 4D heart mesh recovery by building patient-specific meshes from pre-operative full stacks and updating them during procedures using whatever sparse slices are available, down to a single slice.

How does TetHeart handle extreme sparsity and reduce labeling needs?

It incorporates a distillation strategy for accurate reconstruction under severe sparsity and uses weakly supervised motion learning with annotations only at keyframes such as end-diastolic and end-systolic phases.

Jul 16, 2026

Upload by Himbo