Computer Vision News - October 2018

36 Wednesday Daniel Rueckert Daniel, tell us about your work. I lead a group with two of my colleagues, Ben Glocker and Bernhard Kainz, which works really in biomedical image analysis. Our focus is very much on exploiting machine learning to try to acquire medical images, but also interpret medical images in a better way. You had this passion 20 years before it became a fashion. Exactly... Well, actually, I started really working on image registration on, what we call, classical medical image analysis. Then, over the years, we really developed an interest in machine learning techniques, very much the classical techniques originally. Of course, for a number of years, we’re very much also interested in deep learning techniques, which have taken the community by storm. How did you become interested in this subject many, many years before it came into fashion? There’s not always a strategy involved. I think you also need some degree of luck that you’re picking a subject which is promising. Actually, I picked it because I was very much interested in it. I came into the medical imaging community as a pure computer scientist by training. I always had an interest in machine learning techniques. I learned everything I know about neural networks during my undergraduate studies in the late 80s, early 90s when the subject was very unpopular because actually, we couldn’t really solve any real-world problems. I remember my first experiments with neural networks were limited to networks with 200 neurons that would run for hours and hours and hours. Last week, I interviewed Alyosha Efros and he told me: “ When I started, nobody cared about vision, because it did not work .” [ both laugh ] Exactly! I can certainly attest to that. I have first-hand experience of that as well. Do you tell your students that there is a component of luck in the story? Not always, I think, especially in the Daniel Rueckert is a Professor of Visual Information Processing as well as the Head of the Department of Computing at Imperial College London. His research group is interested in developing novel, computational techniques for the analysis of biomedical images. “ My first experiments with neural networks were limited to networks with 200 neurons that would run for hours and hours and hours… ”