Depth-aware template tracking for robust patient motion compensation for interventional 2-D/3-D image fusion

In interventional radiology, pre-operative three dimensional (3-D) images (e.g. CT and MRI) can be fused to the real-time 2-D fluoroscopic images, known as 2-D/3-D image fusion or overlay. 2-D/3-D registration is usually performed at the beginning to ensure an accurate overlay. However, the misalignment caused by patient motion has to be corrected during the procedure. In this paper, we propose an adapted template-based tracking approach that fits well into the novel depth-layer-based framework, where 3-D motion can be estimated by depth-aware tracking in 2-D X-ray images. Templates are generated according to the 2-D/3-D matching procedure, where the 2-D features are matched with the depth layer images generated from the 3-D volume. Our template update strategy takes both initial and current frame into account, which effectively enhances the accuracy of the template tracking. Two strategies are applied to enhance robustness against external disturbances: template lock and motion estimation feedback. The experiment results show that our new approach is capable of estimating 3-D motion and much more robust against external disturbances compared to the Kanade-Lucas-Tomasi tracker that is previously used in the depth-layer-based framework.