Computer Vision News - August 2022

30 Best Paper MIDL computer vision tasks, you go from image to label, whereas inmost medical imaging tasks, at least within reconstruction, you go from image to image. This framework can be modeled agnostic to other image- to-image tasks. As long as you root your augmentation in the physics of the process, this sort of framework canwork. ” If you want to know the secret to their success so that you can create a winning paper yourself, Arjun and Beliz tell us they have been fortunate to be surrounded by talented folk and that it is all about choosing the right team. Try to pick people be prepared for everything, so how can we use that data while integrating our physics knowledge into the model? That’s where the noise and motion modeling comes in. ” Taking home the Best Paper award at MIDL is no mean feat, but the fact that the VORTEX framework is rooted in the physics of how imaging is done could be what most impressed the judges. Traditionally, machine learning has relied upon collecting millions of examples and training a model. This work is driven by the process of how images are acquired. It is focused on MRI, but you can imagine a scenario where it could be generalized to CT or ultrasound. “ The medical field is often limited by the amount of publicly available data that we can perform experiments on, or even if we’re just trying to build the best model, the amount of available labeled data that we have, ” Arjun notes. “ This is going back to oneof the limitationsof standardsupervised methods in deep learning. A takeaway for theMIDL community and beyond is that it’s important to consider how deep learning can be motivated or almost hybridized with the standard analytical and signal processing fundamentals we have known for centuries! That was probably one of the pieces that appealed to the judges. ” Beliz adds: “ Medical imaging and computer vision go hand in hand. In many “ Medical imaging and computer vision go hand in hand.”