MICCAI 2020 Daily - Tuesday

The papers being presented at the workshop cover a variety of imaging modalities, including MRI, X-ray, PET, and CT. MR image reconstruction using deep learning is a hot topic right now and the team anticipated many submissions in this area. At previous events, accelerated MRI has been a key trend . However, this year sees a branching out of the technology to different types of image enhancements and application-specific approaches for MRI and other modalities. “For me, what is exciting about applying machine learning to reconstruction is it allows us to revisit requirements and constraints for sampling and hardware that have always been there in the past,” Patricia tells us . “With accelerated imaging and image enhancements we can push the limits of the diagnostic value that we are getting from these images.” A defining aspect of this workshop is its focus on utilizing machine learning for the image information. Reconstruction does not work with a zero-knowledge assumption. It is a co-existence of classical mathematical models and deep learning parts . “One thing I find interesting is that people are now looking into the acquisition of data as well,” T obias points out. “Not only reconstructing the hidden parameters from a fixed- measurement model, but also learning how to improve the acquisition itself. We can’t directly observe the things we want to reconstruct, so how do we do supervised learning ? What do we train against? These are difficult questions.” 3 MLMIR 17 by Alan Wang, PhD student at Cornell University DAILY Tu e s d a y