A fast registration from 3D CT images to 2D X-ray images

The space location of lung cancer lesions can be obtained by using 2D/3D registration from 3D CT images to dynamic X-ray images. Existing registration methods consumes too much time on calculation, which exceeded acceptable surgical operation time limit. To overcome this problem, we proposed a method based on threshold segmentation to accelerate registration process. By using a fast DRRs algorithm based on Bresenham method, time for generating digitally reconstructed radiographs can be reduced. Threshold segmentation algorithm is adopted to pre-process a fixed image and extract useful information from only a small part of the image instead of the whole image. The running time decreases because fewer pixels are involved in the calculation of the similarity. The accuracy of the registration result is inevitably decreasing. However, it is still within an acceptable range.