A Thin-Plate Spline Based Multimodal Prostate Registration with Optimal Correspondences

Accurate extraction of prostate biopsy samples during Transectal Ultra Sound (TRUS) guided prostate biopsy is facilitated with the registration of pre-acquired Magnetic Resonance (MR) images with the Ultrasound (US) images. This paper proposes a novel method of generating optimal correspondences to register the MR and US images using Thin-Plate Splines (TPS) transformation. The correspondence generation method exploits the prostate shape geometry in both the modalities and is fully automatic. Normalized Mutual Information (NMI) is employed for the quantitative determination of optimal number of correspondences in terms of maximization of registration similarity. Qualitative registration results, that conform to the NMI measures are also shown for different numbers of correspondences. Shepard's interpolation method is used with the TPS in order to deal with the interpolation error of backward TPS transformation. The accuracy of our method of correspondence generation is qualitatively evaluated in comparison with two intuitive geometric contour sampling methods. An average Dice Similarity Coefficient (DSC) value of 0.97 ± 0.01 for 4 patient datasets is obtained for the TPS registration using our novel method of correspondences.