The effect of reconstruction and acquisition parameters for GRAPPA‐based parallel imaging on the image quality

Abstract

Parallel imaging based on generalized autocalibrating partially parallel acquisitions is widely used in the clinical routine. To date, no detailed analysis has been presented describing the dependence of the image quality on the reconstruction and acquisition parameters such as the number of autocalibration signal (ACS) lines N ACS , the reconstruction kernel size (b x 3 b y ), and the undersampling factor R. To evaluate their influence on the performance of generalized autocalibrating partially parallel acquisitions, two phantom data sets acquired with 12-channel and 32-channel receive coils and three in vivo measurements were analyzed. Reconstruction parameters were systematically varied between R 5 2-4, N ACS 5 4-64, b x 5 1-9, and b y 5 2-10 to characterize their influence on image quality and noise. A main aspect of the analysis was to optimize the parameter set with respect to the effectively achieved net image acceleration. Selecting the undersampling factor R as small as possible for a given net acceleration yielded the best result in a clear majority of cases. For all data sets and coil geometries, the optimal kernel sizes and number of ACS lines were similar for a chosen undersampling factor R. In summary, the number of ACS lines should not be chosen below N ACS 5 In MRI, parallel imaging (1,2) is widely used to accelerate the image acquisition by using multiple receiver coils (3) and data undersampling. For Cartesian sampling of kspace, a fraction of the otherwise acquired phase encoding steps can be omitted by exploiting the spatial encoding capability of the receiver coil array. Generalized autocalibrating partially parallel acquisitions (GRAPPA) (1) is a k-space based, autocalibrated (4) parallel imaging algorithm, which is widely used in clinical routine. The method has proven to achieve robust and accurate results for many applications and several extensions and variants of GRAPPA have been presented in recent years (5-7).The application of parallel imaging with increased undersampling factors results in a decrease of the signalto-noise ratio in the reconstructed image (2). In addition to the inherent properties of parallel imaging, the GRAPPA reconstruction is influenced by several acquisition and reconstruction parameters such as the number of autocalibration signal (ACS) lines, the undersampling factor R, and the kernel geometry. Because of a wide variety of possible configurations, it is not obvious how to choose the optimal parameters and how different parameter combinations will influence the resulting image quality and noise level.To date, most studies investigating parameter optimizations have focused on the reconstruction kernel (8,9) or the number of ACS lines (10). However, a detailed analysis of the interrelationship of all parameters, which consider the number of ACS lines N ACS , the undersampling factor R, and the reconstruction kernel (b x  b y ), has not been presented. In this work, in vivo measurements of different anatomical regions (head, heart, and abdomen) and phantom mea...

Recent advances in parallel imaging for MRI

“…A typical kernel may consist of six source points from each coil, with three in the readout direction and two in the phase encoding direction, centered around a target point. Including more source points tends to improve the reconstruction quality [38] but may require additional calibration data, as discussed below.…”

Neurovascular 4DFlow MRI (Phase Contrast MRA): emerging clinical applications

“…Current implementation varies from site to site and vendor to vendor but follows similar trends. Nearly all implementation of 4DFlow utilize parallel imaging, most commonly direct techniques such as SENSE or GRAPPA [7][8][9] but also with indirect methods such as using localized coils. In the case of Cartesian imaging, this reduces the required scan time by a factor of 2-4; and is typically not sufficient to reduce scan times for clinically relevant coverage and spatial resolution (<1 mm) to feasible scan times.…”

Quantifying the Statistical Impact of GRAPPA in fcMRI Data With a Real-Valued Isomorphism

“…The quality of GRAPPA reconstructed images is known to depend on both the kernel size and choice of A. 35,36 Weights were therefore estimated from full FOV data to form G in (6) and (9) for kernels of size 4 × 1, 2 × 3, and 4 × 5 with = 2 and = 3, to explore the effects that each parameter has on theoretical GRAPPA induced correlations.…”