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.

💡 Research Summary

This paper introduces a novel framework for reconstructing time‑resolved, patient‑specific three‑dimensional (3D) aortic geometries from a limited set of standard cine 2D magnetic resonance imaging (MRI) slices. The authors first built a statistical shape model (SSM) using 3,000 synthetic aortic shapes derived from 26 CT‑based meshes of patients with aortic stenosis. Principal component analysis (PCA) was applied, and the first ten modes—accounting for 98.5 % of total variance—were retained as the global shape basis.

For each subject, six to nine cine 2D slices were acquired perpendicular to the aortic arch, manually segmented, and interpolated to a uniform set of contour points. These 2D contours were embedded into a 3D Cartesian space and used as constraints in a differentiable volumetric mesh optimization. The optimization simultaneously updates (i) global shape parameters (SSM coefficients, rigid translation, rotation, isotropic scaling) and (ii) local deformations modeled by radial basis function (RBF) interpolation with polyharmonic splines. The loss function comprises three terms: (a) a surface‑to‑contour L2 distance, (b) a centerline alignment term that forces the mesh‑derived centerline to match a B‑spline fit of the 2D centroids, and (c) regularization penalties on shape amplitudes, rotation angles, and control‑point displacements. Optimization was performed in PyTorch using the Adam optimizer (learning rate = 0.1) for 300 epochs; each cardiac frame was initialized with the solution from the previous frame to enforce temporal smoothness.

The method was evaluated on 30 subjects (19 healthy volunteers, 11 aortic stenosis patients) scanned at two sites with 1.5 T and 3 T scanners. In a subset of ten subjects, a high‑resolution 4D flow MRI (2.5 mm isotropic, 50 ms temporal resolution) served as the reference. Quantitative comparison showed a mean Dice coefficient of 89.9 ± 1.6 %, Intersection‑over‑Union of 81.7 ± 2.7 %, Hausdorff distance of 7.3 ± 3.3 mm, Chamfer distance of 3.7 ± 0.6 mm, and an average absolute radius error of 0.8 ± 0.6 mm. Notably, accurate reconstructions were achievable with as few as six 2D slices, and an iterative slice‑selection strategy identified the most informative cross‑sections for shape recovery.

Age‑related analysis revealed a progressive decline in radial wall strain: 11.00 × 10⁻² in the young group, 3.74 × 10⁻² in middle‑aged, and 2.89 × 10⁻² in the elderly, reflecting reduced aortic compliance with aging. Patients with aortic stenosis displayed similarly reduced strain compared with age‑matched healthy controls.

The proposed pipeline offers several practical advantages. It leverages routinely acquired cine 2D MRI, avoiding the long acquisition times, high costs, and specialized post‑processing required for 4D flow MRI. Consequently, it enables rapid generation of dynamic 3D aortic meshes suitable for downstream computational tasks such as finite‑element biomechanical simulations, patient‑specific hemodynamic modeling, and surgical planning. Limitations include reliance on a synthetic CT‑derived SSM (which may not capture all pathological variations), exclusion of the aortic root from the model, and the need for pre‑defined optimal slice locations. Future work should focus on (1) extending the SSM with real patient CT/MRI data covering a broader spectrum of pathologies, (2) automating slice‑position selection directly from scout images, (3) validating the approach on larger, multi‑center cohorts with diverse aortic diseases (e.g., aneurysms, dissections), and (4) integrating the reconstruction into real‑time clinical workflows. Overall, the study demonstrates that high‑fidelity, time‑resolved 3D aortic geometry can be reconstructed from a minimal set of standard cine 2D MRI slices, opening new avenues for personalized cardiovascular diagnostics and therapy planning.