Meeting Banner
Abstract #1323

GPU Vs CPU Cluster Reconstruction for Low Latency Iterative Reconstruction of First Pass Stress Cardiac Perfusion with Physiological Stress

Sbastien Roujol1, Tamer A. Basha1, Christophe Schlke1, 2, Martin Buehrer1, 2, Warren J. Manning1, 3, Reza Nezafat1

1Medecine, BIDMC / Harvard Medical School, Boston, MA, United States; 2Institute for Biomedical Engineering, University and ETH Zurich, Zurich, Switzerland; 3Radiology, BIDMC / Harvard Medical School, Boston, MA, United States

In cardiac MR perfusion after physiological stress a significant respiratory motion is encountered and coil sensitivity maps are generally corrupted due to patient motion during exercise. Compressed sensing (CS) is an alternative acceleration technique that enables high acceleration even without exploiting temporal dimension or need for coil maps. However, iterative CS reconstruction is lengthy, performed off-line and is not usually integrated into the workflow of a clinical scan. In this study, a GPU-based workflow and a CPU cluster-based workflow have been developed and compared to accelerate the iterative CS reconstruction and minimize the overall reconstruction latency.

Abstract PDF

Keywords

able accelerated acceleration additionally allows alternating although approaches artifacts assessment biomedical breath card channels choice clinical cluster coil commonly communication compatible complete compute computer computing console cost creation database datasets dedicated developed dimension dynamics either employing enable enables equipped even exercise exploiting external fast faster finally graphical guaranteed hardware healthy highly hold hospital immediately implementation improved initial integrated intensive interaction interface introduce investigated iteration iterations iterative latency lengthy limited limiting long majority manually maps medical memory minimal minimization minimize mitigate motion need network nuclear offloaded optimized overall overhead parallel parallelization parallelized patients performance perfusion pharmacological physical physiologic physiological post preprocessed process processing proposed prospective protocol providing radiology randomly reconstruction regularization required resources scanner school send sensing sensitivity sent shared significantly since slices started step steps storing stress studies subjects subsequent substantial supine sustain system table tamer technologist temporal thereby thread treatment unable unit unpredictable user vast viewing waiting warren workstation yang