Computer Vision News - October 2022

23 Xi Fang everything,” Pingkun adds. “We asked them to visually check whether the method worked well. We did a reader study, compared the results, and it showed very good performance.” As we wrap up, James is keen to emphasize the importance of this project being a true collaboration between the medical and the technical side. “Any new technology you present in the medical application base of conferences needs to have a true medical indication,” he asserts. “I have seen so many papers presented at MICCAI that have no real medical indication, meaning you are finding a problem where a problem never existed. We are solving actual medical problems that we encounter during our clinical practice. That is the goal.” This chimes with us - Nassir Navab told us some time ago how important it is to him that his students are always in touch with clinicians and working on solutions driven by what the field needs. James agrees. “Exactly. I know Nassir very well, and he is absolutely correct!” decode it locally, using the just-learned correspondence matrix to transform the local bony movement into the local facial change. “Using the correspondence matrix, we can transform the corresponding bony movement into the corresponding facial change, and then decode it locally so that we canmake theoutput the samedimension as the input bony movement,” Xi explains. “We simulate the facial change based on two key parts – the spatial correspondence between local bone and facial structures and the non-linear relationship between the corresponding bony movement and the facial change.” The team performed a prediction accuracy evaluation using clinical data, comparing their proposed method with two other approaches, including state-of-the-art FEM with realistic lip sliding effect, against which it achieved comparable quantitative accuracy with significantly improved efficiency. “We also asked clinicians to look at the results because numbers do not mean F I