Speaker: Florian Knoll, New York University School of Medicine
Location: Warren Weaver Hall 1302
Date: November 20, 2015, 11:30 a.m.
Host: Dennis Shasha
Magnetic Resonance (MR) image reconstruction can be modeled as an inverse problem. It has been shown during the last 5 years that l1 constrained optimization has a high potential for these types of problems. In most currently used clinical applications the optimiziation problems are convex and solutions can be obtained with first order methods. Practical challenges are mainly related to computational complexity and parameter selection. Upcoming developments like multi-modality imaging and approaches that try to model the underlying physics of the MR signal can violate the assumption of convexity and result in challenging numerical problems.
The first part of this talk will provide a short introduction to the current state of the art of MR image reconstruction. In the second part challenges and open questions of current methods will be discussed from a basic science and an applications point of view. Connections to the fields of parallel computing, machine learning and computer vision will be given.
Refreshments will be offered starting 15 minutes prior to the scheduled start of the talk.