Model-based Postoperative Modeling of Stent-Based Devices from CT : Application to TAVI

As cardiac minimally invasive interventions are progressively substituting conventional surgical procedures, a constantly increasing number of stent-base devices are being implanted. Hence, tools for postoperative assessment and monitoring, which can provide a high-level of information about in-vivo structural and mechanical characteristics of devices, are becoming essential. We propose a novel method to automatically extract a model of stent-based implants from 3D CT volumes that combines data-driven machine learning methods with physicallly geometrical constraints. The Marginal Space Learning framework is used to estimate the position of a stent from an input cardiac image. A robust detector is introduced, which localizes stent struts crossing from an unfolded volumetric representation parameterized by the local coordinate system of the detected device. The model of a stent-frame is determined by computing its realistic deformation subject to internal forces that emulate mechanical behavior. The method was evaluated on post-operative CT volumes of 28 patients that received CoreValve devices during TAVI procedures. Results demonstrated a speed of 10.2 second per volume and average accuracy of 1.27 mm.