Computer Vision News - February 2019

17 are optimizing a specific metric. Obviously, the metrics we use to assess are not necessarily what we care about for clinical purposes. There is an interesting part that we don’t know, and that’s something that I would like to explore in a future challenge. We really hope that the next challenge will happen very soon, maybe not this year, but maybe the year after. We still haven’t conversed on that. The idea would be how would we redesign the challenge in a way that is fairer where most teams have access to the same computational power, where the models themselves can be deployed and trained offsite. We have, for example, containerization of algorithms so that multiple teams can actually submit their models to be trained onsite in the same computational infrastructure. Also, how can we design a challenge that is more clinically relevant? Something where the outcome of the challenge are metrics that are relevant from the point-of-view of the task that is being asked from a clinical point of view. How can we be sure that it’s a statistical methodology that can put together multiple tasks with different metrics? Every task is going to have a metric which is appropriate for what you are trying to solve. Challenge Computer Vision News Medical Segmentation Decathlon