MICCAI 2020

2 Paper Presentation 1 DAILY Mo n d a y it is a worm, the videos we work with have non-linear motion, so we need to model that based on non-linear modelling tools . That is one of the challenges we have been dealing with. Another challenge is a lack of training data . In many other cases, you would have substantial training data that you can use for building neural networks or machine learning tools, but in this case, it is a fairly recent technique. Figuring out how to combine tools and techniques from different topics was pretty challenging.” In the segmentation paper, the team model the image of the C. elegans as a Gaussianmixturemodel ,whichhasbeen used in this context for segmentation Ibefore, but part of the novelty of this work is that it combines the literature of Gaussian mixture models with optimal transport, which is a more recent topic possessing optimization techniques that people are using to build efficient computational models . It could be used as a general method for doing inference in the Gaussian mixture models. The team see this as their contribution to the field of statistics. “Our imaging set-ups and computer vision algorithms are fairly new and that is why we had to build a lot of these tools from the ground up,” Erdem tells us. “C.elegans, especially the type of worms that we are using, have only been introduced in the last five years. There are five or so groups now working on similar problems, but the number isslowly but surely increasing. We hope to be a pioneer in this area in terms of developing this first set of techniques to motivate others to expand on them and improve on our results.” The pipeline the team are building allows scientists to automate many things they have been doing manually . Before, they