Semi-automated quantification of aortic arch length and stiffness in CMR using a 3D cylindrical active surface

KACHENOURA N; ROSENBAUM D; REDHEUIL A; DIETENBECK T; CRAIEM D

Washington

Conferencia; SCMR 20th anual scientific sessions; 2017

Society for Cardiovascular Magnetic Resonance

Background:Aortic stiffness is a well-recognized predictor of cardiovascular risk and is commonly assessed through aortic pulse wave velocity (PWV). In cardiovascular magnetic resonance imaging (CMR), aortic PWV estimation requires the measurement of aortic arch length which is often performed in multiple 2D axial and sagittal oblique views to take into account the complex aortic arch geometry.Methods:We studied 77 individuals (40 men and 37 women, age 50±12y): 23 healthy volunteers and 54 patients with essential hypertension who underwent 3D CMR of the aorta and aortic phase contrast imaging to estimate ascending to descending aorta transit time. The proposed method segments the thoracic aorta volume using explicit active contours where the aortic surface is defined in a cylindrical coordinate system. The computation of the aortic centerline and the definition of aortic anatomical segments required manual initialization of 7 anatomical landmarks from the sino-tubular junction to the celiac trunk. Evaluation of the method was performed by comparing the measured aortic arch length against manual measurements, obtained from the combined analysis of sagittal oblique and axial planes covering the aortic arch. The PWV calculated using both the manual and automatic aortic length measurements were also compared against carotid-femoral PWV (cf-PWV) measured with applanation tonometry.Results:Mean aortic length using our method vs. manual segmentation was 130.5±19mm vs. 112.9±17mm in the control group and 139.6±20mm vs. 120.9±23mm in the hypertensive group. Our method was shown to correlate well with the manual reference in terms of arch length (r = 0.75, p < 0.001; Bland-Altman (BA) mean bias [limits]: 9.5mm [-20.3; 39.2]) and PWV (r = 0.96, p < 0.001; BA mean bias [limits]: 0.58m/s [-1.56; 2.72]). Equivalent correlations of aortic PWV with cf-PWV were found for both aortic length methods (automated: r = 0.52, p < 0.001; manual: r = 0.57, p < 0.001). The mean analysis time was around 5min (2-3min to load the data and setting the landmarks and 2min for the centerline extraction and aortic segmentation) for a Matlab implementation of the software.Conclusion:The proposed method enables fast and semi-automated segmentation of the aorta from a single CMR 3D dataset, providing its centerline length for the estimation of aortic PWV. Such measurements were comparable to the manual aortic length and PWV measurements, which required a time consuming definition of the centerline. Furthermore, such volumetric approach will ultimately enable the extraction of advanced morphological indices such as segmental volumes, arch tortuosity and aortic tapering for further evaluation of aortic physiological changes.