Scalable parallel reconstruction algorithm for magnetic resonance images

Magnetic resonance images (MRI) are critical medical data in noninvasively diagnosing and following disease. The technique uses nuclear magnetic resonance to generate 3-D detailed images of a part or a whole of a human body. The raw data of MRI consists of series of complex numbers that represent phases and amplitudes of signals. The reconstruction procedure of the MRI requires lengthy computational time and sequence of images due to the heavy discrete Fourier transformation. A scalable parallel algorithm, developed for the cluster of heterogeneous computers is presented. This algorithm is simple but effective on a network of workstations, which has a relatively slow speed. Granularity and scalability are considered for determining the best performance. It provides almost a linear speed-up and reasonable runtime that is acceptable within limits.