Computer Vision News - June 2020

2 Summary The L b of the Month 30 He is encouraged by a growing trend for openness, but laments the fact that not everyone is on board with it yet: “We can’t wait to share our code and can’t wait to share our paper and put it on arXiv. I see it as a positive thing, but our field is still lagging behind. Particularly in industry. If you talk to people, they are pretty open, but you have no idea what their five-year plan is. I understand industrial secrets, but as an academic researcher, I would like to know what they are working on so that we don’t crossover. It would make the whole thing more collectively efficient. In making contributions to translational stages, that openness and sharing of information is so important.” He offers a word of caution for people using deep learning technology in medical image computing and surgical interventions. “Quite a few people are inexperienced when it comes to deep learning. We are all guilty in that department,” he says modestly. “Fancy algorithms and promising results are all well and good, but without proper methodology, design and analysis , they won’t translate easily into clinical impact.” However, he is pleased to see more groups are looking at ways to validate their methods using statistical principles with real-world data and prospective patient studies . “ Clinical trials are still a gold standard for determining how useful something is clinically. There is no question about that,” he affirms. “If you have a deep learning algorithm, instead of using retrospective data from other clinical trials, you should start your own clinical trial to validate it. But if you look around the world, how many are driven by a computer scientist? Almost none.” Although Yipeng is optimistic that the tide is starting to change, and he hopes that in the future, academic research will lead the way. “We are impartial and well placed without any other conflicting interests.” Yipeng is hopeful that we will all make it to Lima, Peru in October for MICCAI 2020 , where he will be co-organizing the Advances in Simplifying Medical Ultrasound (ASMUS) workshop with Alison Noble fromUniversity of Oxford and Stephen Aylward from Kitware. Stephen previously organized the Point-Of-Care Ultrasound (POCUS) workshop and it is an extension of that. If international travel is still difficult, he promises us it will be virtual. Ultrasound is a particularly interesting modality right now, but it has its challenges, so research institutes often opt for using other methods. Yipeng tells us this creates a bias which makes it easier to access well-curated data if you are working with neuroimaging . “That’s