REAL-TIME DYNAMIC MRI RECONSTRUCTION: ACCELERATING COMPRESSED SENSING ON GRAPHICAL PROCESSOR UNIT

摘要

The aim of this paper is to propose techniques for realtimedynamic MRI reconstruction from partially sampledK-space measurements. Previous techniques in this areaare either fast but inaccurate or are slow (therefore notamenable for real-time reconstruction) but with higher degreeof accuracy. Recently a Compressed Sensing basedalgorithm has been proposed which yields high degree ofreconstruction accuracy at near real-time speeds. But anyattempt to improve one (speed or accuracy) results in reductionof the other. In this work, we propose to improvespeed and accuracy of the existing CS algorithm by parallelizingit on a Graphical Processing Unit (GPU). Wesee that our parallelized dynamic MRI reconstruction algorithmachieves real-time reconstruction speeds and improvesthe reconstruction accuracy at the same time.