Browse Prior Art Database

METHODS TO ACCELERATE AUTOCALIBRATION FOR COMPRESSED SENSING AND PARALLEL IMAGING

IP.com Disclosure Number: IPCOM000241151D
Publication Date: 2015-Mar-31
Document File: 6 page(s) / 787K

Publishing Venue

The IP.com Prior Art Database

Abstract

Methods of improving the performance of MR image calibration using highly efficient computational methods without any visible change to the image quality are disclosed. An optimally shaped neighborhood of target and source points of undersampled MR image can give a significant improvement in the computational performance without actually affecting the image quality. The methods proposed will enable implementation on the existing MRI devices for high channel count volumes, high quality image reconstruction within clinically acceptable times.

This text was extracted from a Microsoft Word document.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 52% of the total text.

METHODS TO ACCELERATE AUTOCALIBRATION FOR COMPRESSED SENSING AND PARALLEL IMAGING

BACKGROUND

The acquisition of image data in Magnetic Resonance Imaging (MRI) is often slow. Computation of naive and straightforward implementation of the calibration for higher channel count will take several hours on current MR equipment. One approach to reduce the acquisition times is the reconstruction of undersampled data, i.e. to acquire fewer samples that those needed for standard application and to reconstruct the unknown samples using mathematical algorithms. The computational reconstruction techniques to decrease the data acquisition times for sampling data are based on compressed sensing (CS) and parallel imaging (PI), obtaining identical MRI as those obtained with full sampling. With computation of the kernel weights as part of calibration in compressed sensing (CS) and parallel imaging (PI), algorithms such as ESPIRiT, ARC, are computationally expensive, especially for high channel count reconstructions. Further, calculating the inverse for calibration is an expensive operation (order of N3) which impacts the performance significantly. In the conventional method, the problem has been addressed by reusing redundant compute and reducing memory footprint. But, this method uses expensive hardware. Therefore, there is a need for improved computing platforms to provide for reasonable computation times that are clinically practical without increasing hardware costs.

BRIEF DESCRIPTION OF DRAWINGS

Methods to improve performance of calibration in MR imaging without any visible change to the image quality is disclosed, as briefly described with reference to the drawings, in which:

FIG.1 shows a schematic of the different kernel shapes among the target point and source points of neighborhood.

FIG.2 shows a schematic of the algorithmic (ESPIRIT) outputs using different GMAT kernel shapes.

FIG.3 and FIG.4 show the similarities of the image data with coil neighbors and without coil neighbors using ESPIRiT.


DETAILED DESCRIPTION

The methods of optimizing MRI computations disclosed here are described more completely with reference to the drawings in Figure 1, 2, 3 and 4. Figure 1 shows a schematic of the different kernel shapes for GMAT and the number of sets of source points in the neighborhood of the target point across 8 coils. The MRI coils acquire image data from a scanning region of interest, where the data is undersampled. The undersampled data are used to reconstruct the complete image using mathematical algorithms such as ESPIRiT, ARC, or other known methods. The kernel weights are computed by obtaining a least squares fit for predicting target points in the calibration region using a set of source points in their neighborhood. The distance between these points defines the quality of fit and the computational complexity of the solver. The present invention discloses an optimally shaped neighborhood between points that can give a significant improvement in th...