A novel 3D/2D registration method is proposed which is based on robustly matching 3D preinterventional image gradients and coarsly reconstructed 3D gradients from intrainterventional 2D images. The gradient correspondences are established in an iterative process, combining the RANSAC algorithm and a gradient matching cost function. The method was evaluated using 3D CBCT, CT and MR images and 2D X-ray images of to spine segments, validation criteria and metrics. Results show that accuracy and robustness of the method outperforms state-of-the-art 3D/2D registration methods.
COBISS.SI-ID: 6611540
We present a protocol for evaluation of similarty measures for non-rigid registration. The evaluation is based on five intuitive properties that characterize the behaviour of a similarity measure. Properties are estimated locally from similarity measure values that correspond to a range of systematic local free-form deformations. Global similarity measure properties are obtained by combining local properties over image regions or the entire image. The feasibility of the proposed protocol is demonstrated on three similarity measures applied to a number of MR and CT images.
COBISS.SI-ID: 6318164