Computer Vision News - August 2020

Best Paper Award 14 techniques, 10x or 12x is not out of the question , and that opens up new possibilities for dynamic MRI . V alentina explains: “For everything that you want to observe dynamically, such as seeing the organ while it’s moving, you need to do it with the very highest acceleration factor. It’s possible that we will push the boundaries so much that we will create a new use case for which the acceleration is needed in a way that the transition will be smoother.” A key benefit to accelerating MRI reconstruction is reduced cost . The scan time of an MRI directly impacts on the cost and on the possible throughput of patients. If you could perform an MRI in 30 seconds it would cost the sameas an X-ray or an ultrasound and the opportunities would be huge. You could run one for every patient. It could be used as a screening tool for early detection. You could carry out an MRI every threemonths for someoneat riskof cancer. You could perform an MRI every 15 days and track disease perfectly . With incredible advancements in MRI acquisition, Valentina predicts the technology to transform their success into a real-world clinical setting could be just a couple of years away. However, it could take longer to convince the radiology community of its effectiveness. “That’s a completely different ballgame,” she tells us. “It’s on us from the technology perspective to include them in all phases of the development to make them trust the technology in a way that this transition can be smooth.” Besides writing and supervising successful papers, what else do the team do in their day jobs? Valentina’s lab is all about deep learning for musculoskeletal imaging: “It’s half on the acquisition portion, so making the acquisition faster and optimization in acquisition and processing, but also in understanding the disease trajectory. We work a lot on osteoarthritis and degenerative joint diseases.” Francesco has been working mostly with segmentation problems: “Most apply to the problem of task-based image reconstruction. We have this accelerated MRI andtry to learntasksonthat. Coupled with anomaly detection and other secondary tasks to the reconstruction.” Victor is enjoying life at Berkeley and his work in Valentina’s lab: “I’m more on the method side. Trying to not only come up with new methods but come up with methods that are interpretable to understand the underlying causes of the disease.” Best of MIDL 2020