Browse Prior Art Database

GRAPHICS PROCESSING UNIT BASED AUTOCALIBRATING RECONSTRUCTION FOR CARTESIAN SAMPLING (ARC) SYNTHESIS FOR MR PARALLEL IMAGING

IP.com Disclosure Number: IPCOM000206289D
Publication Date: 2011-Apr-18
Document File: 6 page(s) / 40K

Publishing Venue

The IP.com Prior Art Database

Abstract

A technique to redesign an autocalibrating reconstruction for Cartesian sampling (ARC) algorithm for graphics processing unit (GPU) implementation without changing the underlying mathematics that characterizes the algorithm is disclosed. A careful innovative data partitioning is applied to redesign the algorithm and speeds up the algorithm on a hybrid system of CPU and GPU.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 50% of the total text.

Page 01 of 6

RP13587

     GRAPHICS PROCESSING UNIT BASED AUTOCALIBRATING RECONSTRUCTION FOR CARTESIAN SAMPLING (ARC) SYNTHESIS FOR MR PARALLEL IMAGING

BRIEF ABSTRACT

    A technique to redesign an autocalibrating reconstruction for Cartesian sampling (ARC) algorithm for graphics processing unit (GPU) implementation without changing the underlying mathematics that characterizes the algorithm is disclosed. A careful innovative data partitioning is applied to redesign the algorithm and speeds up the algorithm on a hybrid system of CPU and GPU.

     KEYWORDS

    Auto calibrating reconstruction for Cartesian sampling, ARC, GPU, graphics processing unit, CPU, control processing unit(s), data partition, algorithm redesign, CPU-GPU, CPU and GPU, hybrid system, magnetic resonance, MR and image reconstruction.

DETAILED DESCRIPTION

    In general, magnetic resonance (MR) image reconstruction is a process of converting acquired k-space signal data to images. In principle, the process involves performance of Fourier transformations of volume data in all the directions (x, y, and z) in a specific fashion. Several algorithms exist (for example, (autocalibrating reconstruction for Cartesian sampling) ARC, Asset and etc.,) that reduces scan time or collects lesser amount of data and apply mathematical functions along with Fourier transformations to generate a final image.

1


Page 02 of 6

RP13587

    Auto calibrating reconstruction for Cartesian sampling (ARC) is a parallel imaging technique for magnetic resonance imaging (MRI). ARC synthesis is a computationally intensive process, and the existing algorithm and implementation have been developed for a system with a few CPU cores. Accordingly, the time taken grows non-linearly with number of channels; hence the reconstruction lag grows to an unacceptable level (for example, beyond 5 minutes). A run of the ARC synthesis algorithm takes a minimum of 2 to 3 minutes on an 8-core central processing unit (CPU) based system.

    Attempts have been done to reduce the time taken with increased channels and also to reduce the reconstruction lag. One such attempt is, implementing the ARC synthesis algorithm on CPU and parallelization for multiple cores in a workstation, which is not fruitful in results.

    Therefore, there exists a need in the art for developing a technique that overcomes the above listed shortcomings of ARC.

    The present technique enables to redesign the parallel imaging technique such as, ARC algorithm for graphics processing unit (GPU) implementation without changing the underlying mathematics that characterizes the algorithm. In other words, evaluates the use of GPU as a co-processor to perform MR image reconstruction. Implementing the whole of ARC synthesis on the GPU and achieves the increased performance on the GPU. Therefore, a run of the ARC synthesis algorithm that takes 2-3 minutes on an 8-core CPU based system, takes less than 10 seconds on a hybrid system with a CPU and GPU.

    The hybrid system of CPU and GPU is enabled for ARC...