Abstract
Aortic distensibility is a vital indicator of cardiovascular health, yet its manual measurement from cine MR images is labor-intensive and prone to variability. This paper introduces an automated approach using spatio-temporal non-rigid registration. By modeling a 2D+time image sequence as a 3D volume and applying B-spline free-form deformations, the method calculates the Jacobian determinant of the resulting transformation to estimate local area changes. Evaluated on 60 datasets from 20 subjects, the technique achieved an average error of 2.3% compared to manual expert segmentation while reducing processing time from one hour to five minutes.
Introduction
The elasticity of the aorta reflects the cumulative effects of aging and vascular diseases like atherosclerosis. Clinically, this is quantified as "distensibility"—the relative change in the vessel's cross-sectional area per unit of pulse pressure. Traditionally, clinicians must manually trace aortic boundaries across 20–40 frames per heartbeat, a process that is both time-consuming and subject to inter-observer inconsistency. While previous automated models existed, they often lacked the temporal coherence necessary for stable measurements across the full cardiac cycle.
Methodology
The proposed method shifts the focus from frame-by-frame segmentation to global sequence registration. Instead of detecting edges in every image, the algorithm determines a mathematical transformation $T$ that maps a reference baseline frame to every other frame in the sequence.
Spatio-Temporal Non-Rigid Registration
The 2D image sequence is treated as a 3D volume (2D space + 1D time). Using 3D B-splines for Free-Form Deformations (FFD), the algorithm aligns all time frames simultaneously. This spatio-temporal approach ensures that deformations are smooth in both spatial and temporal dimensions, preventing the "jitter" common in frame-by-frame methods.
Area Estimation via Jacobian Determinants
Once the deformation field is established, the local change in area is computed using the determinant of the Jacobian matrix. A Jacobian value greater than 1 indicates expansion, while less than 1 indicates contraction. By integrating these values over the aorta's initial segmented area in the baseline frame, the cross-sectional area is automatically derived for the entire cardiac cycle.
Results and Discussion
The study utilized 60 datasets from 20 subjects, divided into "young" (23–34 years) and "older" (57–80 years) cohorts. Images were captured at three levels of the ascending aorta: proximal, mid, and distal.
The automated method demonstrated remarkable accuracy, with a mean error of only 2.3% relative to manual segmentation. More importantly, the processing time was slashed from approximately 60 minutes to just 5 minutes per section. The tool effectively differentiated age groups: younger subjects exhibited significantly higher expansion ratios (up to 160%), whereas older subjects showed reduced distensibility, typically staying below 120% expansion.
Conclusion
This registration-based approach provides a robust and efficient alternative to manual aortic segmentation. By leveraging temporal coherence and the Jacobian determinant, it offers a precise tool for clinical cardiovascular assessment. Future work will focus on fully automating the initial baseline segmentation and extending the method to more complex vascular structures like the aortic arch.
Summary updated