NASCI AHA Young Investigator Finalists 2012

Abstract Category: CT Aorta/peripheral vascular disease Purpose: Cardiac CT is increasingly used for the assessment of aortic annulus dimensions and orientation prior to transcatheter aortic valve replacement (TAVR), but manual assessment is cumbersome and time-consuming. We evaluated the accuracy and time-effectiveness of semi-automated model-based annulus computation compared to manual planimetry. Methods : Retrospectively ECG-gated dual-source cardiac CT data of 53 consecutive TAVR candidates with severe aortic stenosis (mean age 82.7±7.9 years) were included. Data were reconstructed at 300ms past the R-peak and were analyzed using an automatic 3D aortic valve model. This model is fitted to CT data by discriminative learning methods and incremental search. The model contains the surface of the aortic root, the commissures and hinges and allows for automated morphologic identification of the aortic annulus plane, defined by the most basal hinge points of the aortic cusp. The encompassed aortic annulus contour is delineated based on gray-scales, with manual correction. Manual planimetric measurement using multiplanar reformations was used as the reference standard. Analysis times for both methods were recorded. Data were analyzed using linear regression and Bland Altman plots. Hypothetical prosthesis sizing (23mm prosthesis for 25mm) was compared using κ-statistics. Results : Aortic valve hinge-points were correctly identified in 43/53 patients (83%). Mean effective annulus diameter was 23.1±2.3mm by model-based analysis and 22.9±2.4mm by manual assessment (p=n.s.). Excellent correlation was found between both methods (r=0.98, p<0.01) and Bland Altman analysis revealed no systematic bias. Agreement for prosthesis sizing was found in 39/43 patients (κ=0.85). Mean analysis time was significantly (p<0.001) reduced for model-based measurements (26±8sec), compared with manual planimetry (98±12sec). Conclusion: Semi-automated morphologic aortic annulus quantification derived from an aortic valve model enables fast and accurate procedural planning in excellent agreement with manual planimetry and has the potential to improve cardiac imaging workflow in the evaluation of patients prior to TAVR.