A Latent ODE Approach to Spatiotemporal Modeling of Cine Cardiac MRI

Cardiac magnetic resonance imaging (CMR) captures rich spatiotemporal information about ventricular structure and motion, but conventional risk models use only a few image-derived indices from selected cardiac phases. We present a latent dynamical model that encodes bi-ventricular anatomy and full-cycle cine motion as a continuous latent trajectory, using heart-rate-aware neural ordinary differential equation (ODE) dynamics and a graph-based mesh autoencoder to reconstruct anatomically consistent 3D+t ventricular motion. A covariate-conditioned prior defines the expected end-diastolic latent state, and a Cox proportional hazards model tests whether deviations from this prior predict incident heart failure. We studied 72,386 UK Biobank participants without baseline cardiovascular disease, including 367 incident heart failure events. In a held-out evaluation subset, adding the latent score to refitted pooled cohort equations improved the stratified C-index from 0.704 to 0.785, compared with 0.764 for seven established cardiac markers. Compared with non-graph and non-ODE approaches, the proposed model gave the best trade-off between reconstruction fidelity, generative realism, and downstream prognostic performance. These results suggest that continuous full-cycle modeling of ventricular motion provides informative cardiac phenotypes beyond conventional CMR summaries, while external validation in more representative patient cohorts is required before clinical risk-prediction use.

Temporally resolved aortic 3D shape reconstruction from a limited number of cine 2D MRI slices

Background and Objective: We propose a shape reconstruction framework to generate time-resolved, patient-specific 3D aortic geometries from a limited number of standard cine 2D magnetic resonance imaging (MRI) acquisitions. A statistical shape model of the aorta is coupled with differentiable volumetric mesh optimization to obtain personalized aortic meshes. Methods: The statistical shape model was constructed from retrospective data and optimized 2D slice placements along the aortic arch were identified. Cine 2D MRI slices were then acquired in 30 subjects (19 volunteers, 11 aortic stenosis patients). After manual segmentation, time-resolved aortic models were generated via differentiable volumetric mesh optimization to derive vessel shape features, centerline parameters, and radial wall strain. In 10 subjects, additional 4D flow MRI was acquired to compare peak-systolic shapes. Results: Anatomically accurate aortic geometries were obtained from as few as six cine 2D MRI slices, achieving a mean +/- standard deviation Dice score of (89.9 +/- 1.6) %, Intersection over Union of (81.7 +/- 2.7) %, Hausdorff distance of (7.3 +/- 3.3) mm, and Chamfer distance of (3.7 +/- 0.6) mm relative to 4D flow MRI. The mean absolute radius error was (0.8 +/- 0.6) mm. Significant age-related differences were observed for all shape features, including radial strain, which decreased progressively ((11.00 +/- 3.11) x 10-2 vs. (3.74 +/- 1.25) x 10-2 vs. (2.89 +/- 0.87) x 10-2 for young, mid-age, and elderly groups). Conclusion: The proposed method enables efficient extraction of time-resolved 3D aortic meshes from limited sets of standard cine 2D MRI acquisitions, suitable for computational shape and strain analysis.