Search for a command to run...
Image-guided procedures have become routine in medicine. Due to the nature of three-dimensional (3-D) structure of the target organs, two-dimensional (2-D) image acquisition is gradually being replaced by 3-D imaging. Specifically in the diagnosis of prostate cancer, biopsy can be performed using 3-D transrectal ultrasound (TRUS) image guidance. Because prostatic cancers are multifocal, it is crucial to accurately guide biopsy needles towards planned targets. Further the gland tends to move due to external physical disturbances, discomfort introduced by the procedure or intrinsic peristalsis. As a result the exact position of the gland must be rapidly updated so as to correspond with the originally acquired 3-D TRUS volume prior to biopsy planning. A graphics processing unit (GPU) is used in this study to compute rapid updates performing 3-D motion compensation via registration of the live 2-D image and the acquired 3-D TRUS volume. The parallel computational framework on the GPU is exploited resulting in mean compute times of 0.46 seconds for updating the position of a live 2-D buffer image containing 91,000 pixels. A 2x sub-sampling resulted in a further improvement to 0.19 seconds. With the increase in GPU multiprocessors and sub-sampling, we observe that real time motion compensation can be achieved.