UBC Theses and Dissertations
Deformable prostate registration from MR and TRUS images using surface error driven FEM models Taquee, Farheen
TransRectal Ultrasound (TRUS) is used for image guidance during prostate biopsy and for treatment planning of brachytherapy due to low cost and accessibility in operating room. However, tumors have better visibility in Magnetic Resonance (MR) images. The fusion of TRUS and MR images of the prostate can aid with the diagnosis and treatment planning for prostate cancer and with post-brachytheraphy quality assurance. We developed a 3D deformable registration method using the segmentations obtained from TRUS and MR images and a biomechanical model that employs stiffness values derived from elastography. The segmented source volume is meshed and a linear finite element model is created for it. This volume is deformed to the target image volume by applying surface forces computed by assuming a negative relative pressure between the non-overlapping and the overlapping regions of the volumes. This pressure drives the model to increase the volume overlap until the surfaces are aligned. We tested our algorithm on prostate surfaces extracted from postoperative MR and TRUS images for 14 patients and pre-operative MR and TRUS images for 4 patients, using a model with elasticity within the range reported in the literature for the prostate. We used three evaluation metrics for validation: the Dice Similarity Coefficient (DSC) (ideally equal to 1.0), the volume change in source surface during registration, and the Target Registration error (TRE) defined as the mean distance between landmarks such as urethrae and calcifications. For post-operative images, we obtained a DSC of 0.96±0.02 and a TRE of 1.5±1.4mm. The change in the volume of the source surface was 1.5±1.4%. For pre-operative images, we obtained the DSC of 0.96±0.01 and a TRE of 1.3±0.8mm. The change in the volume of the source surface was -0.9±0.2%. Our results show that this method is a promising tool for physically-based deformable surface registration. We also used our technique to register ultrasound strain images to free mount histo-pathology images with the goal of correlating cancer with areas of low strain. This was done using relative stiffness values derived from vibroelastography data. We also performed Computed Tomography (CT) and Ultrasound (US) kidney surface registration using this technique.