Computer Vision News - May 2020

2 Summary Research 4 Hi everyone! In the hope and wish that you all keep well and in good health and shape, this month you're going to read a review about a Deep Learning approach in Magnetic Resonance Imaging (MRI) and especially on the aspect efficiency during the acquisition and reconstruction of the signals and image respectively. The importanceofMRI ismajor in clinical practicebecauseof various reasons, including the fact that it offers excellent soft-tissue contrast. How about high-resolution volumetric images though? This is possible but with a caveat: speed. The acquisition process tends to be very slow. This is negative for the patients and depending on the imaged organ, may result in discomfort (e.g. holding the breath for long durations). A state-of-the-art solution is Compressed Sensing (CS), introduced by Michael Lustig in John M. Pauly's group at Stanford University. Compressed Sensing allows much faster acquisitions, therefore reduced scan times, by undersampling. That introduces a new problem though: slower reconstruction times. The reason is that it leads to an ill-posed linear inverse reconstruction problem. How is that solved? By incorporating prior information, by sparsity regularisation in a proper transform domain or finite differences and Total Variation (TV). Those algorithms and approaches are iterative, which introduces an intensity in the used resource, thus longer processing times. The proposed solution for those problems is a reconstruction using historic patient data. The main concept behind it is that it's possible to train a network that learns Every month, Computer Vision News reviews a research paper from our field. For May, we chose to talk about a GAN MRI paper: Deep Generative Adversarial Networks for Compressed Sensing Automates MRI. We are indebted to the authors (Morteza Mardani, Enhao Gong, Joseph Y. Cheng, Shreyas S. Vasanawala, Greg Zaharchuk, Lei Xing, and John M. Pauly), for allowing us to use their images to illustrate this review and in particular Morteza for additional images from a presentation of the paper. You can find their paper at this link. Deep Generative Adversarial Neural Networks for Compressive Sensing (GANCS) MRI, a review. by Ioannis Valasakis