Computer Vision News - September 2020

2 Guest 56 Best of ECCV 2020 is the most extraordinary step forward that you would like to see in the next couple of years in medical imaging, done by our community of scientists? I can tell you something which I would like to see happen, which I think would be very exciting. We hear a lot about artificial intelligence, deep learning, medical imaging, and so on. As you know, there are different schools of thought, ranging from “This will save the world” , to “Nahhh” . [ laughs ] So I would like, in the next few years, to see peace, from the technical point of view. [ laughs ] This would mean to reconcile deep learning techniques with conventional techniques, from one important point of view: explainability. If somehow we crack this and write systems that can actually explain the techniques, that to me will definitely be a game changer! Is that happening now? What would bring us to that? It's not happening as I described it. There's a lot of excellent work around the world which is moving quickly and you have to be careful in judging this, because there are colleagues around the world who like to write, “We're going to solve the problem tomorrow” . You have to take this in perspective, look at the whole issue and see where they are, discuss with doctors and so on. One thing I've learned, and I will certainly recommend to junior people, is that you have to make an effort not to talk only to people like you. In our field, it is absolutely super vital, essential, to keep an open mind and realize that clinical people, people who we write the software for - patient organizations, legislative bodies, governing bodies - each read this in a different way and have different priorities. Nassir Navab told me he wants all of his engineers to keep a continuous communication with clinicians and surgeons. Yes. This is happening to a certain extent in the UK, for instance with dedicated DTC, or doctoral training centers. For medical things, there's an emphasis on interdisciplinarity, and so you brew the next generation of scientists with this open mind, ready for the dialog and to gain some rather precise understanding of how doctors reason and what the doctors want. Which is typically a classic situation with engineering, where, some time ago it was: “Give me 20 images. I willwrite an algorithm… Problem solved!” It doesn't work like that. [ both laugh ] What R&D advancement are you seeing on the industry side? Like companies making large investments in ophthalmic R&D? Absolutely. I have two things that spring to mind immediately. One is equipment. And the other is, again, software, image analysis,andsoon. Intermsofequipment, if you look back to just 10 years ago, the situation has taken leaps forward. Take

RkJQdWJsaXNoZXIy NTc3NzU=