[home] [Personal Program] [Help]

---

Session: Heart
Session starts: Thursday 26 January, 10:30
Presentation starts: 11:45
Room: Room 558

---

Sabine Verstraeten (Eindhoven University of Technology)
Martijn Hoeijmakers (ANSYS )
Frans van de Vosse (Eindhoven University of Technology)
Wouter Huberts (Eindhoven University of Technology)

Abstract:
In silico clinical trials are a promising method to increase the efficiency of the development of transcatheter aortic valve implantation (TAVI) devices. With an in silico trial, devices can be tested on virtual patients by computer models. Each patient of the virtual cohort is represented by a synthetic aortic valve geometry. Two important aortic valve morphologies to include are: (1) the shape of the left ventricular outflow tract (LVOT), either convergent or divergent, and (2) the angle between the LVOT and the ascending aorta (∠LVOT-AA). These morphologies influence the occurrence of complications, such as conduction problems [1], and paravalvular leakage [2]. To the best of our knowledge a framework to generate synthetic aortic valve geometries that considers these morphologies is not yet available. Therefore, the aim of this research is to develop a framework to generate synthetic aortic valve geometries, that (1) are physiologically plausible, and (2) allow for selection of the aforementioned morphologies. Non-parametric statistical shape modeling (SSM) [3, 4] was used to extract the mean shape and shape variance (shape modes) from a set of 97 stenotic aortic valve geometries. Each geometry within or outside this data set was approximated by adding a weighted combination of 24 shape modes to the mean shape. With the SSM 500 synthetic geometries were generated by sampling new weight combinations from an inferred distribution [5, 6]. Logistic regression and linear regression models were used to filter synthetic geometries on LVOT morphology and ∠LVOT-AA respectively. A non-parametric multivariate ANOVA test revealed that the 500 synthetic geometries did not differ significantly from the set of 97 real patient geometries (p = 0.47 > 0.05). The LVOT shape filter and the ∠LVOT-AA filter successfully selected the aforementioned morphologies with a sensitivity of 97% and 94% respectively. These results demonstrate that the framework developed in this study, (1) succeeded in generating synthetic geometries that are physiologically plausible, and (2) makes it possible to select geometries with certain morphologies. Consequently, this framework has the potential to generate synthetic data sets for in silico TAVI trials.