Communicore Room C1-17
Gainesville, FL 32611
Alireza Entezari, Ph.D.
Computer & Information Science & Engineering Department,
University of Florida
Imaging from low-dose projection data has been the holy grail of X-ray CT research for several decades. Modern algorithms enable high-resolution reconstruction from low-dose data by incorporating statistics and, more recently, learning and neural-network-based priors into the reconstruction process. Despite promising dose-reduction results from these iterative reconstruction algorithms, the classical Filtered Back Projection (FBP) algorithm and its variants remain widely employed in commercial scanners. The main impediment in translating these modern reconstruction algorithms to practice has been their computational cost (i.e., long reconstruction time and space complexity) compared to classical methods. Recent work from our group introduced a radically different approach for computational challenges in iterative CT reconstruction by establishing connections to techniques previously developed in scientific computing. This previously unexplored nexus provides an opportunity to revamp and highly accelerate the optics computations involved in forward and back-projection — essential ingredients in all modern iterative reconstruction algorithms. The talk will provide a high level presentation of these ideas and show resulting improvements in the performance of low-dose CT reconstruction both from speed and accuracy viewpoints, compared to state-of-the-art, patented, methods in this area.
Bio: Dr. Alireza Entezari is an associate professor in the Computer & Information Science & Engineering Department at the University of Florida. His research interests include mathematical signal processing, sampling theory and optimization with applications in imaging and visualization.
Dr. Christine E. Schmidt