Non-rigid 2D-3D registration using anisotropic error ellipsoids to account for projection uncertainties during aortic surgery.

Overlay of preoperative images is increasingly being used to aid complex endovascular aortic repair and is obtained by rigid 2D-3D registration of 3D preoperative (CT) and 2D intraoperative (X-ray) data. However, for tortuous aortas large non-rigid deformations occur, thus a non-rigid registration must be performed to enable an accurate overlay. This article proposes the use of Thin-Plate Splines (TPS) to perform non-rigid 2D-3D registration. Intraoperative X-ray data contain no spatial information along the X-ray projection direction. Our approach accounts for this lack of spatial information by the use of an approximating TPS with non-isotropic error ellipsoids, where the major ellipsoid axis is aligned with the X-ray projection direction. Experiments are carried out using 1D-2D and 2D-3D simulated data and 2D-3D interventional data. Simulated results show that our proposed method is 1.5 times more accurate than interpolating TPS based registration. Interventional data results show how large rigid registration errors of 9mm can be reduced to 4mm using our proposed method.