Organizations: Department of Electrical and Computer Engineering, Johns Hopkins University, Baltimore, MD, USA · Gordon Center for Medical Imaging, Department of Radiology, Massachusetts General Hospital and Harvard Medical School, Boston, MA, USA · Center for Advanced Medical Computing and Analysis, Department of Radiology, Massachusetts General Hospital and Harvard Medical School, Boston, MA, USA
Cine cardiovascular magnetic resonance (CMR) analysis relies on multi-frame sequences capturing the full cardiac cycle. However, standard multi-frame acquisition depends heavily on electrocardiogram (ECG) gating and repeated breath-holds, posing challenges in uncooperative populations, resource-limited settings, and temporally corrupted datasets. Existing methods that synthesize full cardiac sequences either rely on explicit ECG signals to parameterize myocardium function, or employ deformable registration without physiological constraints, failing to faithfully reproduce clinically relevant dynamic metrics such as ejection fraction (EF) and ventricular contraction magnitude. We present PhaseFlow, a unified generative framework that overcomes both limitations. PhaseFlow estimates a non-linear cardiac phase signal directly from the input sequence via a segmentation-derived left-ventricular (LV) area curve, capturing the asymmetric dynamics of systole and diastole without any ECG dependency. At inference, this phase signal is provided by a pathology-specific template, informing phase-specific frame generation. A rectified flow model conditioned on the phase and slice position synthesizes the full cardiac motion trajectory in the latent space, decoded into a diffeomorphic displacement field that warps end-diastole pixel intensities directly, eliminating the reconstruction blur often accompanying the variational autoencoder. On the ACDC benchmark, PhaseFlow achieves superior physiological fidelity and image realism, with best LV volume curve R2, structural similarity (SSIM) and generative quality (FID) among all baselines. Ablation studies confirm that each proposed component contributes measurably to the overall performance.
Figures & tables
Figure 1: PhaseFlow encodes the ED frame into latent features, predicts a phase-conditioned latent velocity via rectified flow, and decodes a diffeomorphic displacement field that warps ED pixels into a fully synthesized cine sequence.
Figure 2: Linear phase (left) scatters ES across ϕ values; nonlinear phase (right) aligns ES at ϕ≈π for different pathological types (e.g., normal and dilated cardiomyopathy).
Image Quality
Physiological Fidelity
Method
PSNR ↑
SSIM ↑
LPIPS ↓
FID ↓
EF MAE ↓
Vol Corr ↑
Vol R2↑
Vol MAE ↓
ED Repeat
30.63
0.950
0.022
18.22
51.37
—
− 2.050
0.245
ConvLSTM
27.57
0.903
0.137
79.94
48.12
0.082
− 1.539
0.230
Direct Reg
31.89
0.955
0.031
17.26
19.66
0.924
− 0.105
0.116
CVAE
30.89
0.949
0.037
26.04
22.29
0.921
− 0.038
0.122
EchoDiff ( Phi et al. 2024 )
15.20
0.219
0.391
115.17
15.47
0.414
− 1.000
0.211
Table 1: Main results on ACDC test set (30 patients). Best in bold , second-best underlined .
Image Quality
Physiological Fidelity
Variant
PSNR ↑
SSIM ↑
LPIPS ↓
FID ↓
EF MAE ↓
Vol Corr ↑
Vol R2↑
Vol MAE ↓
Full (pathology phase)
31.72
0.956
0.023
12.72
17.79
0.867
0.363
0.101
A1: VAE decoder
27.28
0.882
0.128
89.63
20.22
0.888
− 0.548
0.128
A2: Linear phase
31.48
0.954
0.024
12.91
21.30
0.827
0.069
0.114
A3: No LNCC
15.17
0.238
0.176
50.25
14.88
0.876
0.295
0.094
A4: No vol loss
31.76
0.955
0.023
11.43
26.80
0.869
0.071
0.133
Table 2: Ablation study on the ACDC test set. Each row removes one component from the full PhaseFlow model.
Figure 3: LV volume curves. Ground-truth (solid) vs. predicted (dashed) trajectories per pathology group.
Figure 4: Qualitative comparison (HCM and NOR). Rows 1–3: GT, PhaseFlow-VAE, and our PhaseFlow (Deform) at ED, mid-systole, ES, and mid-diastole. Row 4: predicted displacement magnitude. PhaseFlow produces sharp, anatomically coherent frames; PhaseFlow-VAE shows characteristic reconstruction blur.
Image Quality
Physiological Fidelity
Setting
PSNR ↑
SSIM ↑
LPIPS ↓
FID ↓
EF MAE ↓
Vol Corr ↑
Vol R2↑
Vol MAE ↓
ACDC (in-domain)
31.72
0.956
0.023
12.72
17.79
0.867
0.363
0.101
M&Ms (zero-shot)
32.80
0.937
0.016
0 4.82
19.89
0.845
0.309
0.114
Table 3: In-domain vs. zero-shot generalization. PhaseFlow trained on ACDC and evaluated on M&Ms without any adaptation.
Appendix figures & tables7 assets
Supplementary material from the paper’s appendix.
Appendix
Figure 5: Per-pathology nonlinear phase templates (coloured) vs. linear baseline (grey dashed). All groups show a fast-then-slow profile: rapid systolic rise, then slower diastolic progression.
Figure 6: Per-frame phase RMSE: nonlinear group template (solid) vs. linear baseline (dashed). Shaded region: error gap. Nonlinear template reduces approximation error by up to 1.7× .
Figure 7: det(Jφ) distribution (S&S, K=7 ). Left : global histogram; red dashed: fold boundary ( det=0 ). Right : per-pathology mean ± std. Zero folding across all pixels confirms diffeomorphic synthesis.
Metric
Mean ± Std
Image quality (per slice, n=298 )
PSNR (dB)
31.72±3.33
SSIM
0.956±0.019
LPIPS
0.023±0.012
Physiological fidelity (per patient, n=30 )
EF MAE (%)
17.79±11.15
Appendix
Table 4: Per-case mean ± standard deviation of PhaseFlow on the ACDC test set. Image metrics are per-slice ( n=298 ); physiological metrics are per-patient ( n=30 ).
Wilcoxon p (PhaseFlow vs.)
Metric
ED Rep.
ConvLSTM
CVAE
Dir. Reg.
EF MAE
1.9×10−9
1.9×10−9
2.0×10−2
0.36
Vol Corr
—
5.6×10−9
0.36
0.72
Vol R2
1.9×10−9
6.2×10−8
2.3×10−4
1.6×10−2
Vol MAE
5.6×10−9
2.6×10−8
2.3×10−4
0.064
Appendix
Table 5: Paired Wilcoxon signed-rank test ( n=30 patients) against four baselines. Every cell is a p -value; smaller means stronger evidence that PhaseFlow is better . Bold = significant at α=0.05 . PhaseFlow is significantly better on Vol R2 against all baselines; non-bold cells (Vol Corr vs. CVAE/Dir. Reg., EF/Vol MAE vs. Dir. Reg.) indicate no significant difference. Vol Corr vs. ED Repeat is undefined (constant volume curve).
Quantity
Train ( n=1048 )
Val ( n=143 )
σ of z1
0.506
0.538
σ of anchor Evae(repeat(⋅))
0.503
0.535
σ of z1−z0
0.171
0.174
α/σ(z1)
19.8%
18.6%
α/σ(z1−z0)
58.3%
57.4%
Appendix
Table 6: Empirical latent statistics ( σ = per-element standard deviation). The two splits agree closely, indicating the latent scale is a stable property of the frozen encoder rather than a split artifact.
Figure 8: Per-pathology qualitative results (one representative patient per group). Top row : ground truth; bottom row : PhaseFlow synthesis. ED frame ( t=0 , leftmost) is the sole model input.
Cine cardiovascular magnetic resonance (CMR) captures the cardiac cycle as a four-dimensional (4D) sequence, but standard acquisition requires electrocardiogram (ECG) gating and repeated breath holds. Visual realism alone does not establish accurate patient-specific ejection fraction (EF) or ventricular volumes. We present PhaseFlow3D, a generative framework that synthesizes a complete 4D cine sequence from a single end-diastolic (ED) three-dimensional (3D) volume without ECG. To capture asymmetric systolic and diastolic dynamics, it represents the cardiac cycle as a piecewise linear phase anchored at ED and end-systolic (ES) time points. At inference, a population-level canonical template supplies this phase without patient-specific temporal information. A phase-conditioned rectified flow model generates a cardiac motion trajectory in latent space. Radial Contraction Decomposition converts each latent state into a 3D displacement field, combining a physics-informed radial component for centripetal myocardial contraction with an image-conditioned residual for rotation and out-of-plane motion. Each frame is generated by directly warping the ED volume, bypassing variational autoencoder decoding. On the combined ACDC and M&Ms benchmark, PhaseFlow3D achieves the lowest EF mean absolute error, the only positive left-ventricular volume-curve R2, and the best distributional quality among compared methods. Ablations confirm each component's contribution. Downstream evaluations demonstrate the utility of the synthesized sequences and displacement fields for segmentation, pathology classification, label propagation, and myocardial strain analysis.
Shiyi Wang, Ruochen Sun, Xiang Li +2
Department of Electrical and Computer Engineering, Johns Hopkins University, Baltimore, MD, USA · Gordon Center for Medical Imaging, Department of Radiology, Massachusetts General Hospital and Harvard Medical School, Boston, MA, USA · Center for Advanced Medical Computing and Analysis, Department of Radiology, Massachusetts General Hospital and Harvard Medical School, Boston, MA, USA
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.
Xiaocheng Fang, Zhengyao Ding, Guangkun Nie +9
State Key Laboratory of General Artificial Intelligence, Peking University · School of Intelligence Science and Technology, Peking University, Beijing, China · Polytechnic Institute of Zhejiang University, Zhejiang University, Hangzhou, China +4
Cardiac motion over a cardiac cycle is crucial for quantifying regional function and is strongly affected by cardiovascular diseases. Since temporally dense mesh sequences are difficult to obtain in practice, we focus on leveraging the more accessible end-diastolic frame to infer a full-cycle sequence. Due to strong regional and disease-specific differences, traditional methods often oversmooth the data by relying on generative models that are optimized for global patterns. To address this problem, we propose Region-Aware and Phenotype-Adaptive Bi-Ventricular Cardiac Motion Synthesis (RePCM) for single frame Bi-ventricular mesh motion completion. In Stage I, a reconstruction network learns vertex wise motion descriptors and clustering yields a data driven functional partition, providing an explicit motion derived region structure. In Stage II, a Region-Specific Injection Module enforces masked, synchronized region exchange within a conditional VAE, preserving localized specific dynamics and restricting cross-region mixing. A Phenotype-Adaptive Mixture-of-Experts prior conditioned on ED shape uses anatomy-guided cues to model latent motion trends and capture inter-disease variability. Experiments on three datasets covering different cardiovascular diseases show consistent gains in geometric and functional metrics and improved preservation of region specific dynamics.
Xuan Yang, Xiaohan Yuan, Hao Li +3
School of Biomedical Engineering, National University of Singapore, Singapore · School of Automation, Southeast University, Nanjing, China · School of Computer Science and Technology, Nanjing University of Aeronautics and Astronautics, Nanjing, China +1