ECCV 2020 Daily - Wednesday

The CV4MI tutorial was given from the perspective of industrial research scientists (Alison Q. O’Neil and Keith A. Goatman) from Canon Medical Research Europe , a Canon R&D centre based in Edinburgh. After opening with a guest introduction from Prof. Keith W. Muir, a consultant stroke physician based in Glasgow, on How AI will transform healthcare: A clinician’s view , Keith then followed with an Introduction to medical imaging and spoke about the Challenges in medical image analysis , before Alison illustrated with some Case studies in deployment of medical AI and discussed the multimodal state of the art Beyond imaging: Multimodal AI using the whole patient record before outlining more generally Future opportunities in medical image analysis . Sunday’s one-hour sessions were lively forums for questions and discussion ranging from “What techniques do you use to cope with limited data in medical imaging?” [Answer: transfer learning (somewhat), domain knowledge (a lot, right through the ML pipeline), techniques for learning with unlabelled data (more andmore)] to “What is the difference between explainability and interpretability for medical imaging? ” [Answer: Strictly speaking, explainabilitymight be howwell we can explain a network’s predictions to a clinician, whereas interpretability might be how well we as computer scientists can understand the network predictions – these are often synonymous] to “How can I train model for medical prognosis prediction?” [Answer: Good question! Prediction is hard… See our talk on multimodal medical AI for more]. Below we outline the case for computer vision in medical image analysis. All videos for the tutorial are available on the online platform . For a personal challenge, try the medical imaging trivia questions and Medical Imaging Bingo within our “Introduction to Medical Imaging” presentation! The Need Health services worldwide are struggling to maintain service levels while costs 2 Tutorial 18 DAILY W e d n e s d a y “Computer Vision for Medical Imaging” (CV4MI) by Alison Q. O’Neil and Keith A. Goatman