Computer Vision News - June 2019

36 Computer Vision News Women in Science during the intervention. They use tactile feedback. They listen to the acoustic signals. They see something, not on the screen, but on the real patient. All of this data is not being captured yet. There is a lot of variation in the procedure. It’s not one 3D image that we’re talking about. It’s hours of surgery. It’s multiple signals that are not being captured. It’s really complex. In the end, I don’t think we really have success stories, yet. This is one of the things that I would like to discuss at the workshop. It seems that you and your colleagues will be able to do great things soon, as a result of your research. You still sound very humble and do not give your research much credit. How can you connect these two things? I don’t like overselling because I want people to know they can trust my words. I will not say it really works until the point when I’m 100% certain about it. I would like to ask you now about the Medical Decathlon. I know that you were involved in this challenge, which was sponsored by DeepMind, NVIDIA, and RSIP Vision. Our readers know about it because we interviewed Michela Antonelli and Jorge Cardoso a few months ago. What are your key take home thoughts? I think that the outcome was really interesting. I was recruited to the organizing team due to my team’s expertise in challenges in general. We analyzed a lot of challenges, how rankings are performed, and what the problems are. Based on all my experience that I had before, I was very afraid that our ranking would result in many shared places such that we would not be able to make any strong conclusions. However, what we found was quite interesting. The winning algorithm had the hypothesis that you don’t need to invent a new net architecture, but you just need to know how to use existing architectures and how to properly train your algorithm. It’s an entirely new hypothesis. The title of the paper was actually quite provocative, which is “ No New Net ” (nnU-net because it builds upon the widely used U-net proposed by Olaf Ronneberger and colleagues, 2015). It’s not about inventing something entirely novel, but it’s about doing things right. I like this hypothesis. I was really surprised and excited to see that the algorithm won by a large margin. I have to say though that I have a kind of a conflict of interest here because the winning team is part of my husband’s group [ laughs ]. The algorithm is open- source though and was uploaded as a docker container, so there was definitely no cheating involved [ laughs again ]. Women in Science “ The backbone of my work. Amazing team with great team spirit! ”