MICCAI 2021 Daily – Thursday

is very smooth, so tends to provide less information. When Yamid joined, the team were already working on the liver, but he had the idea to improve the accuracy of the augmented reality by using multiple views . Merging the information coming from the different views is challenging. The liver deforms between views, so it is difficult to find an accurate registration that respects the constraints for each of these views. A special deformable registration approach is needed to achieve accurate augmented reality. “ Finding this relationship between the images was the most difficult task, ” he tells us. “ First, we tried to merge contour information from all the views. For every image we marked 2D landmarks on the liver surface in order to deform the preoperative 3D models. If there was texture information available in the form of keypoint correspondences, then we included it to constrain the registration. If you don’t have keypoint correspondences, then you can just use the contour information and merge them together to obtain a final registration. ” In the end, the team developed methods for two scenarios . The first scenario is for where the liver does not deform between images. For this, they assume the liver is rigid across the image and use it to constrain the registration. They have two methods that work like this. The first method uses the contours without the texture information. By computing an average model across all the different deformations, they are able to obtain 9 DAILY MICCAI Thursday Yamid Espinel