Image registration for prostate MR guided biopsy using automated biomechanical modeling

PURPOSE To investigate the effect of extending a non-rigid surface-based registration method with biomechanical modeling for prostate MR guided biopsies on the target registration error (TRE) using internal reference landmarks. The method is fully automated and we compare accuracy to previous results obtained with manual optimization of parameters in every patient. METHOD AND MATERIALS The accuracy of a novel non-rigid registration method involving biomechanical modeling to account for deformations inside the prostate was determined. While MR-TRUS registration is the ultimate goal, we used MR guided MR biopsy imaging data from six consecutive patients for this evaluation. The data included T2-weighted images (0.8x0.8x3.0 mm) before and after insertion of a needle guide causing deformation of the prostate. The needle guide had an orientation and dimension comparable to a transrectal ultrasound (TRUS) probe. The prostate in the two images was segmented and corresponding surface meshes were generated in both images by assuming identical prostate orientations. Next, a tetrahedral volume mesh was generated from the image before needle insertion. Prostate deformations due to needle insertion were simulated using the surface displacements as boundary condition. A 3D thin-plate spline deformation field was calculated by registering the mesh vertices. The TRE was defined as the Euclidean distance between registered and reference landmark position and was calculated for 45 reference landmarks manually annotated in both T2-weighted images. The results of this automated method were also compared to previous results obtained with manual optimization. RESULTS The median TRE of the automated surface-based registration method with biomechanical regularization was 2.21 mm (range 0.55-7.32 mm), which was significantly lower than a median TRE of 3.02 mm (range 0.85-7.95 mm) obtained without biomechanical regularization (P<0.01). The median TRE of the automated method is higher than the previous result of 1.88 mm, but not significantly different (P=0.10). CONCLUSION Non-rigid surface-based image registration extended with biomechanical modeling can be automated and improves the registration accuracy for prostate MR guided biopsies. CLINICAL RELEVANCE/APPLICATION The automated surface-based registration method extended with biomechanical modeling is applicable to MR-TRUS registration and can help to improve effectiveness of MR guided TRUS biopsy procedures.