Work

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

Year: 2026

Type: article

Abstract: Background and objective To assess the feasibility and accuracy of reconstructing time-resolved, three-dimensional, subject-specific aortic geometries from a limited number of standard cine 2D magnetic resonance imaging (MRI) acquisitions. This is achieved by coupling a statistical shape model with a differentiable volumetric mesh optimization algorithm. Methods Cine 2D MRI slices were manually segmented and used to reconstruct subject-specific aortic geometries via a differentiable mesh optimization algorithm, constrained by a statistical shape model. Optimal slice positioning was first evaluated on synthetic data, followed by in-vivo acquisition in 30 subjects (19 volunteers and 11 aortic stenosis patients). Time-resolved aortic geometries were reconstructed, from which geometric descriptors and radial strain were derived. In a subset of 10 subjects, 4D flow MRI data was acquired to provide volumetric reference for peak-systolic shape comparison. Results Accurate reconstruction was achieved using as few as six cine 2D MRI slices. Agreement with 4D flow MRI reference data yielded a 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. The mean absolute radius error along the aortic arch was (0.8 ± 0.6) mm. Secondary analysis demonstrated significant differences in geometric features and radial strain across age groups, with strain decreasing progressively with age at values of (11.00 ± 3.11) × 10 –2 vs. (3.74 ± 1.25) × 10 –2 vs. (2.89 ± 0.87) × 10 –2 for the young, mid-age, and elderly groups, respectively. Conclusion The proposed framework enables reconstruction of time-resolved, subject-specific aortic geometries from a limited number of standard cine 2D MRI acquisitions, providing a practical basis for downstream computational analysis.

Source: Computer Methods and Programs in Biomedicine

DOI: 10.1016/j.cmpb.2026.109425

Authors: Gloria Wolkerstorfer, Stefano Buoso, Rabea Schlenker, Jochen von Spiczak, Robert Manka, Sebastian Kozerke

Institutions: University of Zurich, ETH Zurich, Institute for Biomedical Engineering, University Hospital Zurich

Language: English

Topic: Medical Image Segmentation Techniques

Subfield: Computer Vision and Pattern Recognition

Field: Computer Science

Domain: Physical Sciences

SDG: Good health and well-being

Open Access status: hybrid

Estimated APC paid: $3,350