Computer Vision News - June 2020

2 Summary The Lab of the Month 6 College and KU Leuven. The team submitted a MICCAI paper around this fetoscopic intervention, where they coupled vessel segmentation with mosaicking from the vessels. It achieved very promising results and is informing what they are doing next. “It’s an area where there are very few published results, so anything is new,” he says enthusiastically. “It is already working well and that’s something very exciting!” One challenge they face is having limited data to work with, due to it being a relatively rare procedure and there being only a set number of videos to draw on. Faced with this, is it still possible to build models that work well? Francisco explains they work with phantom models and simulation to increase data sets . “I think both having a lot of data and having a small amount of data are challenging in different ways,” he points out. “In our case in surgery we deal a lot more with the fact that we don’t have enough data, but I don’t see this as strictly more challenging. Maybe when we have a lot of data we are at a more developed stage, but what we are comparing against is also more developed, so in both cases there are challenges.” Another area of work is gastric surgery , where the team are trying to do 3D reconstruction of the environment. It’s a very dynamic environment, so Francisco says they can’t simply assume they can reconstruct everything. Navigation and localisation of instruments is another important component. Also, localisation and 3D reconstruction for the purpose of better segmentation. What has Francisco seen of other people’s work that has impressed him? “Work around using deep learning