Fast 2-D/3-D registration using a laptop PC with commodity graphics hardware

Image registration is a technique for finding point correspondences between two different images taken at different times and/or in different modalities This technique plays an important role in computer-aided surgery. However, CPU implementations take more than 10 minutes to complete a registration task due to a large amount of computation. Therefore, some acceleration techniques are required to use this technique for surgical assistances, where response time is strictly limited in a short time. One acceleration technique is to use GPUs (graphics processing units) equipped on PC graphics cards, which are rapidly increasing performance. The objective of our work is to reduce registration time by using GPUs. This paper presents our GPU method that accelerates three key procedures of 2-D/3-D rigid registration: digitally reconstructed radiograph generation, neighbour filtering, and reduction operation. We also investigate the usability of our method from the viewpoint of registration time. The experimental results show that our method completes a registration task within 15 seconds, and thus we find that our GPU method is fast enough to use it for surgical assistances.