Purpose:To evaluate an optimized k-t-space related reconstruction method for dynamic magnetic resonance imaging (MRI), a method called PEAK-GRAPPA (Parallel MRI with Extended and Averaged GRAPPA Kernels) is presented which is based on an extended spatiotemporal GRAPPA kernel in combination with temporal averaging of coil weights. Materials and Methods:The PEAK-GRAPPA kernel consists of a uniform geometry with several spatial and temporal source points from acquired k-space lines and several target points from missing k-space lines. In order to improve the quality of coil weight estimation sets of coil weights are averaged over the temporal dimension. Results:The kernel geometry leads to strongly decreased reconstruction times compared to the recently introduced k-t-GRAPPA using different kernel geometries with only one target point per kernel to fit. Improved results were obtained in terms of the root mean square error and the signal-to-noise ratio as demonstrated by in vivo cardiac imaging. Conclusion:Using a uniform kernel geometry for weight estimation with the properties of uncorrelated noise of different acquired timeframes, optimized results were achieved in terms of error level, signal-to-noise ratio, and reconstruction time. DYNAMIC MRI is an important foundation for many clinical applications such as time-resolved (Cine) cardiac imaging for the assessment of left ventricular function. To achieve sufficient spatial and temporal resolution fast data acquisition is essential, particularly for applications that require breath-holding. In order to reduce total acquisition time or to increase spatiotemporal resolution, parallel imaging techniques such as SENSE or GRAPPA have been introduced. By decreasing the number of phase-encoding steps by a reduction factor R, imaging can be substantially accelerated (1,2). Based on varying sensitivity in multiple receiver coil arrays, parallel imaging reconstruction algorithms remove resulting aliasing artifacts either in the image domain (2) or regenerate the missing data in k-space (1). For parallel imaging reconstruction using the kspace related GRAPPA technique the image reconstruction and the combination of images from different receiver coils are decoupled. Therefore, the process of unaliasing of the uncombined coil images (generated from each coil) can be optimized separately.For conventional parallel dynamic MRI using GRAPPA, the central k-space for each timeframe is fully sampled, forming the autocalibration signal (ACS) lines, while the outer k-space is undersampled in the phase-encoding (ky) direction according to a user-defined reduction factor R. All timeframes are reconstructed independently using a kernel with a certain extension in kx-and ky-direction. The kernel is shifted across the ACS lines in order to estimate the coil weights needed for the reconstruction (or interpolation) of the missing lines in outer k-space. For the reconstruction process of the missing k-space lines, the kernel is shifted by an increment of R in ky-direction over the undersampl...

The purpose of this study was to combine a recently introduced spatiotemporal parallel imaging technique, PEAK-GRAPPA (parallel MRI with extended and averaged generalized autocalibrating partially parallel acquisition), with two-dimensional (2D) cine phase-contrast velocity mapping. Phase-contrast MRI was applied to measure the blood flow in the thoracic aorta and the myocardial motion of the left ventricle. To evaluate the performance of different reconstruction methods, fully acquired kspace data sets were used to compare conventional parallel imaging using GRAPPA with reduction factors of R ‫؍‬ 2-6 and PEAK-GRAPPA as well as sliding window reconstruction with reduction factors R ‫؍‬ 2-12 (net acceleration factors up to 5.2). PEAK-GRAPPA reconstruction resulted in improved image quality with considerably reduced artifacts, which was also supported by error analysis. To analyze potential blurring or low-pass filtering effects of spatiotemporal PEAK-GRAPPA, the velocity time courses of aortic flow and myocardial tissue motion were evaluated and compared with conventional image reconstructions. Quantitative comparisons of blood flow velocities and pixel-wise correlation analysis of velocities highlight the potential of PEAK-GRAPPA for highly accelerated dynamic phase-contrast velocity mapping.

Whole-body fat quantities derived noninvasively by using a continuously moving table Dixon acquisition were directly compared with ADP. The accuracy of the method and the high reproducibility of results indicate its potential for clinical applications.

Two-dimensional (2D) axial continuously-moving-table imaging has to deal with artifacts due to gradient nonlinearity and breathing motion, and has to provide the highest scan efficiency. Parallel imaging techniques (e.g., generalized autocalibrating partially parallel acquisition GRAPPA)) are used to reduce such artifacts and avoid ghosting artifacts. The latter occur in T 2 -weighted multi-spin-echo (SE) acquisitions that omit an additional excitation prior to imaging scans for presaturation purposes. Multiple images are reconstructed from subdivisions of a fully sampled k-space data set, each of which is acquired in a single SE train. These images are then averaged. GRAPPA coil weights are estimated without additional measurements. Compared to conventional image reconstruction, inconsistencies between different subsets of k-space induce less artifacts when each k-space part is reconstructed separately and the multiple images are averaged afterwards. These inconsistencies may lead to inaccurate GRAPPA coil weights using the proposed intrinsic GRAPPA calibration. It is shown that aliasing artifacts in single images are canceled out after averaging. Phantom and in vivo studies demonstrate the benefit of the proposed reconstruction scheme for free-breathing axial continuously-moving-

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...

No abstract

Parallel imaging algorithms require precise knowledge about the spatial sensitivity variation of the receiver coils to reconstruct images with full field of view (FOV) from undersampled Fourier encoded data. Sensitivity information must either be given a priori, or estimated from calibration data acquired along with the actual image data. In this study, two approaches are presented, which require very little or no additional data at all for calibration in two-dimensional multislice acquisitions. Instead of additional data, information from spatially adjacent slices is used to estimate coil sensitivity information, thereby increasing the efficiency of parallel imaging. The proposed approaches rely on the assumption that over a small range of slices, coil sensitivities vary smoothly in slice direction. Both methods are implemented as variants of the GRAPPA algorithm. For a given effective acceleration, superior image quality is achieved compared to the conventional GRAPPA method. For the latter calibration lines for coil weight computation must be acquired in addition to the undersampled k-spaces for coil weight computation, thus requiring higher k-space undersampling, that is, a higher reduction factor to achieve the same effective acceleration. The proposed methods are particularly suitable to speed up parallel imaging for clinical applications where the reduction factor is limited to two or three. Magn

In this paper we investigate methods to adapt a system for disfluency removal to different data properties. A gradient descent algorithm for parameter optimization is presented which achieves 85.1% recall and 93.1% precision on the English Verbmobil corpus and 53.0% recall and 79.0% precision on the Mandarin Chinese CallHome corpus. This compares to the results produced with hand-optimization on the test set. Furthermore we investigated the impact of cross-validation and training set selection on recognizer output. Finally we examined speaker dependent disfluency production behavior and clustered training data accordingly in order to improve the overall system.