Robust registration of 3-D ultrasound and CT images of the liver for image-guided intervention

The registration of multi-modal images of the same organ is beneficial in various clinical applications. However, conventional methods usually require a time-consuming and inconvenient manual process for pre-alignment. In this paper, we present an automatic and robust registration algorithm of 3-D B-mode US and CT images of the liver. The proposed algorithm first automatically segments vessels and liver surface from a 3-D B-mode US image by efficiently eliminating unwanted clutters and noise. It then predicts a reliable initial transform parameter set in a non-iterative manner by maximizing the geometric correlation between the skeletons of vessels segmented from both images. Finally, the algorithm refines the obtained parameters iteratively to find the optimal affine transform parameters by jointly using vessel and liver surface information. Experimental results for 20 clinical datasets show that the proposed algorithm successfully registers a 3-D B-mode US image to its corresponding 3-D CT image even with a large misalignment.