ISBI Daily - Saturday

test set until one has fully developed the technique on the training set, with limited evaluation on a validation set. That’s very important as well. The idea is a lot of these experiments separate the data into training validation and test sets, where the test set is the final assessment of generalizability of the technique – and that’s what we want, generalizability. We want to know that the technique is going to work outside of the training and validation datasets. It’s very important to have data hygiene to not use that test set until after the training and validation are completed. Do you see this happening? You know, it may be happening. It’s not entirely clear from many of the presentations whether that actually is happening. I don’t hear many of the speakers actually saying it. They talk about separate training validation and test sets, but it’s not always clear how many rounds of training and validation and testing were done. If data hygiene was respected - to use your term. Right. How do you see this community evolving in the coming years? The size and the rate of public datasets are going to increase, so that’s going to lead to more robust validations of the techniques. I think it will get to the point where if a paper doesn’t do that on an external dataset, then it’ll probably be harder for those researchers to be successful at getting acceptance to meetings like this. My experience has been that the rate of availability of datasets is increasing, and I think it’ll continue to do so. Can you suggest any untapped clinical problems which is clinically relevant? I think some examples include some of the gynecologic malignancies. I tend to see less in that particular area. May I suggest what Anne Carpenter told me? Mental illnesses - this is an area in which our profession might give a much bigger contribution than it has done to date. What do you think of that? I think there’s a lot of truth to that. Particularly the psychiatric illnesses, I think, is an area where there’s a tremendous mental health burden in our society. That’s an area where many contributions could be made. I think that the imaging techniques are going to be disseminating worldwide at a great rate of speed, and that we should also be aware of the needs of not just the major economies, but also the developing economies in which the techniques that we’re working on could be very helpful, in ways that we may not be thinking about. I would encourage people to think about those other under-served communities as well. ISBI DAILY Saturday 19 Ron Summers