Automated 3D aortic strain mapping from 4D CTA showed excellent correlation with manual measurements (R>0.95) and good correlation with CMR (R=0.608-0.765) for assessing aortic stiffness.
Does an automated deep learning and deformable registration method accurately compute 3D aortic strain maps from 4D CTA compared to manual annotation and CMR in patients with bicuspid aortic valves?
An automated deep learning-based method can accurately and reproducibly generate 3D aortic strain maps from 4D CTA, correlating well with manual annotations and CMR.
Tasa de eventos absoluta: 0% vs 0%
Abstract Background/Introduction Arterial stiffness is one of the earliest detectable markers of functional changes in the aortic wall and a powerful predictor of adverse events 1. However, its current assessment requires significant expertise and effort and provides only global or two-dimensional stiffness descriptors, limiting the ability to evaluate local heterogeneity of aortic properties. As deformable image registration applied to serial computed tomography angiographies (CTA) provides accurate 3D aortic growth assessment 2,3, we hypothesize that it would offer reliable 3D aortic strain maps once applied to time-resolved CTA. Purpose To develop and validate an automated method for computing 3D aortic strain maps from four-dimensional (4D) CTA using deep learning (DL) pre-processing and deformable registration. Methods Twenty-nine patients with bicuspid aortic valve (70% male, aged 41 to 78 years) and available multiphase CTA acquisitions were included. Aortic segmentation and anatomical landmarks were automatically obtained at end-diastole via DL algorithms and tracked across the 4D CTA phases using deformable image registration. Additionally, manual expert annotations were obtained in 15 patients for comparison. From the time-resolved aortic meshes, ascending (AAo) and descending (DAo) aorta deformation maps (longitudinal and circumferential strain, and aortic surface relative change (SARC)) were computed. Maximum (systolic to diastolic) deformation was extracted and used for validation by comparing (i) automatic vs manual strain measurements (n=15), (ii) strain dependency with age (n=29), and cross-modality comparison with AAo circumferential (n=22) and longitudinal (n=18) strain by cardiac magnetic resonance. Results Excellent correlation was obtained between automatic and manual strain for all deformation parameters (R0.95, p0.0001). The well-known stiffening of the AAo and DAo (Figure 1) with age was obtained for all parameters and confirmed in the single patient with two available time-resolved CTAs acquired 6 years apart (Figure 2). In cross-modality validation with CMR, 4D CTA-based measurements showed good correlation with CMR for longitudinal strain in the AAo (R=0.608, p=0.007) and circumferential strain in the AAo (R=0.765, p0.001) and DAo (R=0.747, p0.001). Conclusion Accurate and reproducible 3D aortic strain maps can be obtained automatically from 4D CTA via DL and deformable image registration. This approach enables the local assessment of aortic stiffness, thus extending the possibilities in the study of the pathophysiology and prognosis of aortic diseases.Figure 1 Figure 2
Catalá-Santarrufina et al. (Sat,) reported a other. Automated 3D aortic strain mapping from 4D CTA showed excellent correlation with manual measurements (R>0.95) and good correlation with CMR (R=0.608-0.765) for assessing aortic stiffness.