Automatic cardiac motion tracking from cine cardiac magnetic resonance (CMR) imaging is essential for quantifying myocardial motion abnormalities associated with cardiovascular disease. Existing methods fall into two groups: those that represent motion without anatomical correspondence across subjects and therefore limit inter-subject motion assessment; and those that represent motion with globally corresponding points, which typically require a time-consuming and error-prone semi-automatic step to fit a cardiac atlas to the subject space. To address these limitations, this study proposes CardioDrift, a fully automatic, atlas-based probabilistic framework that represents subject-specific cardiac motion on a 3D atlas mesh with globally corresponding points. The Coherent Point Drift (CPD) formulation is generalised to jointly register the atlas surfaces of multiple cardiac regions to the anatomically corresponding contours extracted from cine CMR segmentation maps throughout the cardiac cycle. Slice-shift correction is embedded in the same probabilistic model, estimating slice-wise translations that suppress the misalignment artefact limiting atlas-fitting accuracy. Evaluation on 1301 randomly selected UK Biobank subjects and the M&M dataset, CardioDrift attains Dice similarity coefficients (DSCs) of 92.23±8.70%, 82.11±7.56%, and 88.69±4.13% for the left ventricle (LV), myocardium (Myo), and right ventricle (RV), exceeding the strongest baseline on every distance and overlap metric of both datasets.
Figures & tables
Fig. 1: Overview of CardioDrift. (a) Alignment phase: the atlas Y is fitted to the subject’s cardiac space at ED, which also estimates the per-slice translations. (b) Propagation phase: Y is deformed to the corrected contours Xl at each frame, providing Zl . {R,t,s} are the rigid parameters, tk are the slice shift correction translations, and ϕ is the non-rigid deformation field.
Dataset
Stage
Method
HD95 (mm)
MSD (mm)
DSC (%)
Endo
Epi
RV
Endo
Epi
RV
LV
Myo
RV
UK Biobank
Rigid
CardioDrift
14.18 ± 4.43
16.98 ± 5.06
22.30 ± 4.78
4.50 ± 1.21
4.77 ± 1.39
5.72 ± 1.52
90.72 ± 10.56
70.25 ± 12.27
81.01 ± 9.28
CardioDrift w/o SSC
17.19 ± 6.94
18.93 ± 7.42
21.16 ± 4.83
5.62 ± 2.60
5.80 ± 2.48
4.96 ± 1.16
86.52 ± 11.90
55.76 ± 12.68
86.19 ± 8.82
CPD
21.45 ± 11.30
23.09 ± 11.12
32.66 ± 15.98
7.51 ± 3.88
8.11 ± 4.71
10.29 ± 6.58
68.13 ± 29.96
41.76 ± 27.89
57.96 ± 31.88
ICP
38.10 ± 6.46
36.75 ± 6.28
44.35 ± 6.42
11.00 ± 1.64
10.76 ± 1.74
15.70 ± 2.09
34.19 ± 17.39
11.70 ± 5.01
50.32 ± 5.94
PCRNet
24.79 ± 4.62
25.37 ± 5.08
40.29 ± 6.85
8.50 ± 1.56
7.71 ± 1.82
9.93 ± 1.54
56.01 ± 14.77
25.03 ± 11.18
75.04 ± 5.76
TABLE I: CardioDrift against baseline registration methods on both datasets, with the slice-shift-correction (SSC) ablation. HD95 and MSD (mm) are computed against the subject contour points, DSC (%) against the reference masks. On UK Biobank, the rigid stage is evaluated at ED and the non-rigid stage across all frames; on M&M, the rigid stage is evaluated at ED and the non-rigid stage at ED and ES. Bold : best per column within each stage of each dataset.
Fig. 2: Qualitative results. (a) Rigid atlas fitting for CardioDrift and the four baselines, with the mid-ventricular SAX image, ground-truth mask, and extracted contour points on the left; per method, the fitted atlas over the contour points and the rasterised mask. (b) Motion tracking across the cardiac cycle for CardioDrift, CPD, CPD-Net, and DeepMesh; per method, the propagated atlas and its rasterised mask, against the SAX image and ground-truth mask.
Fig. 3: Slice-shift correction. Atlas fitted (a) with and (b) without correction, over the subject’s contour points. Axial view of the original middle slice SAX image overlaid with the rasterised mask (c) with and (d) without slice shift correction. Coronal (e, f) and Sagittal (g, h) views of the original and corrected SAX images, with ground-truth masks overlaid.
Dataset
Config
Endo
Epi
RV
HD95 (mm)
DSC(%)
Endo
Epi
RV
All Regions
LV
Myo
RV
All Regions
UK Biobank
Full
✓
✓
✓
14.18 ± 4.43
16.98 ± 5.06
22.30 ± 4.78
17.82 ± 3.84
90.72 ± 10.56
70.25 ± 12.27
81.01 ± 9.28
80.66 ± 9.84
Endo only
✓
✗
✗
11.79 ± 3.30
15.84 ± 4.56
37.38 ± 7.93
21.67 ± 4.01
92.59 ± 4.72
72.29 ± 6.43
35.03 ± 10.18
66.64 ± 5.33
Epi only
✗
✓
✗
15.67 ± 4.98
16.33 ± 4.71
39.37 ± 9.56
23.79 ± 5.29
87.54 ± 13.43
66.92 ± 13.33
33.51 ± 10.06
62.66 ± 10.36
RV only
✗
✗
✓
41.00 ± 17.61
41.24 ± 16.52
17.19 ± 3.21
33.14 ± 11.68
55.13 ± 37.03
30.56 ± 20.94
81.46 ± 9.44
55.72 ± 21.86
M&M
Full
✓
✓
✓
17.23 ± 6.98
22.15 ± 7.29
27.00 ± 8.62
22.13 ± 6.65
83.82 ± 20.34
60.21 ± 19.10
75.01 ± 18.54
73.02 ± 18.07
TABLE II: Ablation study of the shape-aware modelling across anatomical regions (rigid stage), evaluated over UK Biobank and M&M subjects. ✓ and ✗ denote whether the shape-aware term is enabled for the corresponding region. All metrics are reported as Mean ± Std. Bold : best per column within each dataset.
Medical image registration aligns scans across time, patients, or imaging modalities to track anatomy and support diagnosis, treatment, and image-guided procedures.
Diffusion-based unsupervised image registration has been explored for cardiac cine MR, but expensive multi-step inference limits practical use. We propose FlowReg, a flow-matching framework in displacement field space that achieves strong registration in as few as two steps and supports further refinement with more steps. FlowReg uses warmup-reflow training: a single-step network first acts as a teacher, then a student learns to refine from arbitrary intermediate states, removing the need for a pre-trained model as in existing methods. An Initial Guess strategy feeds back the model prediction as the next starting point, improving refinement from step two onward. On ACDC and MM2 across six tasks (including cross-dataset generalization), FlowReg outperforms the state of the art on five tasks (+0.6% mean Dice score on average), with the largest gain in the left ventricle (+1.09%), and reduces LVEF estimation error on all six tasks (-2.58 percentage points), using only 0.7% extra parameters and no segmentation labels. Code is available at https://github.com/mathpluscode/FlowReg.
Yunguan Fu, Wenjia Bai, Wen Yan +3
University College London · InstaDeep · Imperial College London
Cardiac cine MRI provides time-resolved visualization of cardiac motion and is essential for evaluating ventricular function and anatomy. In routine clinical acquisition, however, cine sequences are frequently temporally sparse or degraded by arrhythmia, motion, and scan-time constraints, which can hinder reliable quantification and downstream analysis. We present CardioTrace, a motion-aware cine synthesis framework that generates a full cardiac cine sequence from limited input while explicitly modeling and tracing cardiac motion. CardioTrace couples two complementary pathways: a pixel-translation branch that directly predicts cine frames in a residual formulation, and a flow-field branch that estimates time-resolved deformation fields from a reference phase to provide interpretable motion correspondences across frames. To further preserve anatomical plausibility, a segmentation branch supervises clinically meaningful cardiac structures at key phases, and a cross-branch consistency objective encourages the synthesized appearance dynamics to follow the explicit motion guidance, mitigating temporal drift and unrealistic shape transitions over long sequences. The framework also supports prompt-style conditioning to incorporate auxiliary patient or acquisition information, improving robustness across heterogeneous multi-center data. Experiments on public cardiac cine MRI datasets demonstrate that CardioTrace improves perceptual quality and temporal coherence over representative GAN- and diffusion-based baselines, while better preserving ventricular morphology throughout the cardiac cycle. Our codes are available after review.
Zhi Chen, Le Zhang
School of Engineering, College of Engineering and Physical Sciences, University of Birmingham, Birmingham, UK
Accurate 4D whole-heart mesh reconstruction from sparse cine MRI is critical for creating cardiac digital twins, but remains challenging due to limited 2D slice coverage and the complex coupling between cardiac shape and motion. Existing methods often rely on intermediate contour fitting and typically reconstruct static, single-phase, or partial cardiac geometries, limiting their ability to capture full-chamber dynamics. We propose a novel end-to-end framework for reconstructing temporally resolved whole-heart meshes from multi-view 2D cine MRI sequences by learning an image-to-mesh mapping. The framework incorporates a differentiable contour renderer inspired by the Beer-Lambert attenuation principle, enabling anatomy-aware supervision of 3D+t mesh deformation through contour-based projection losses. To improve temporal consistency across the cardiac cycle, we further introduce a multi-scale temporal modeling module that integrates global cycle-level dynamics with local inter-frame coherence to generate smooth and physiologically plausible mesh trajectories. The proposed method achieved a whole-heart mean absolute error of 1.68 ± 0.31 mm and a motion jitter of 0.77 ± 0.17 mm/frame3, outperforming existing methods with lower reconstruction error and substantially improved motion smoothness. It also improved 2D contour alignment across multiple cine MRI views and supported downstream proof-of-concept electrophysiological simulation. The code will be released publicly upon acceptance of the manuscript for publication.
Xiaoyue Liu, Dongcheng Cang, Xiaohan Yuan +3
Department of Biomedical Engineering, National University of Singapore, Singapore · School of Automation, Southeast University, Nanjing, China · Department of Medicine, National University of Singapore, Singapore +1