UBC Theses and Dissertations
Image-based guidance for prostate interventions Khallaghi, Siavash
Prostate biopsy is the gold standard for cancer diagnosis. This procedure is guided using a 2D transrectal ultrasound (TRUS) probe. Unfortunately, early stage tumors are not visible in ultrasound and prostate motion/deformations make targeting challenging. This results in a high number of false negatives and patients are often required to repeat the procedure. Fusion of magnetic resonance images (MRI) into the workspace of a prostate biopsy has the potential to detect tumors invisible in TRUS. This allows the radiologist to better target early stage cancerous lesions. However, due to different body positions and imaging settings, the prostate undergoes motion and deformation between the biopsy coordinate system and the MRI. Furthermore, due to variable probe pressure, the prostate moves and deforms during biopsy as well. This introduces additional targeting errors. A biopsy system that compensates for these sources of error has the potential to improve the targeting accuracy and maintain a 3D record of biopsy locations. The goal of this thesis is to provide the necessary tools to perform freehand MR-TRUS fusion for prostate biopsy using a 3D guidance system. To this end, we have developed two novel surface-based registration methods for incorporating the MRI into the biopsy workspace. The proposed methods are the first methods that are robust to missing surface regions for MR-TRUS fusion (up to 30% missing surface points). We have validated these fusion techniques on 19 biopsy, 10 prostatectomy and 11 brachytherapy patients. In this thesis, we have also developed methods that combine intensitybased information with biomechanical constraints to compensate for prostate motion and deformations during the biopsy. To this end, we have developed a novel 2D-3D registration framework, which was validated on an additional 10 biopsy patients. Our results suggest that accurate 2D-3D registration for freehand biopsy is feasible. The results presented suggest that accurate registration of MR and TRUS data in the presence of partially missing data is feasible. Moreover, we demonstrate that in the presence of variable probe pressure during freehand biopsy, a combination of intensity-based and biomechanically constrained 2D-3D registration can enable accurate alignment of pre-procedure TRUS with 2D real time TRUS images.
Item Citations and Data
Attribution-NonCommercial-NoDerivs 2.5 Canada