V1238: Real-time tracking of biplane TRUS images by image registration accelerated with Graphics Processing

V1238: Real-time tracking of biplane TRUS images by image registration accelerated with Graphics Processing Unit


Introduction and Objectives
To assist precise prostate targeting and accurate 3D mapping of the biopsy trajectory in the prostate model, computer-assisted guidance systems are developed. In previously reported systems, the magnetic or optical sensing devices for tracking of TRUS probe sometimes disturb or constrain the range of the free-hand manipulation, and thus those could affect against smooth and precise biopsy operation. The aim of this study is to develop a novel tracking method of TRUS probe to locate current scanning position in the 3D space of the prostate based on image registration which is accelerated by Graphics Processing Unit (GPU).

The proposed method has two steps. First: reconstruction of a 3D-US image of the prostate from the series of simultaneous biplane US images by the continuous free-hand axial scanning of the entire prostate. Second: registration between the reconstructed 3D US image and live biplane US images to locate current scanning position. Reconstruction accuracy was evaluated by comparing the reconstructed 3D US image with the MR image of a prostate phantom. Tracking accuracy was evaluated by comparing the estimated position and orientation with those measured by the optical tracker.

Our GPU-assisted image registration system allowed real-time demostration of the 3D orientation of bi-plane images, which was overlaid on the initially reconstructed 3D prostate volume data (Figure). Mean surface error of the reconstructed image was 0.71 mm and overlap ratio was 0.962 (1.0 means completely matched). Mean tracking error was 3.5 mm and 3.8 degrees in translation and rotation, respectively. The frame rate was 2.1 frames per second. At the time of performing prostate biopsy guided by the real-time bi-plane probe; since the spatial orientation of the bi-plane TRUS probe could be real-timely tracked, every hyper-echoic needle trajectory on the bi-plane US image could be registered onto the initially constructed 3D prostate model.

We developed the new tracking method of biplane TRUS images, allowing free-hand manipulation of the TRUS probe similar to routine TRUS-biopsy practice. This technology enabled us to document 3D location of each biopsy trajectory in the reconstructed 3D prostate image, without use of attachment of any external tracking devices and 3D-TRUS probe.

Funding: none