Theory And Reconstruction Method Of Parallel Magnetic Resonance Spectroscopic Imaging

Magnetic resonance spectroscopic imaging (MRSI) is formed through an organic combination of the space information provided by MRI and the spectroscopic information by MRS, which can get the spectrum in multiple voxels and the metabolites map for observing the metabolite states. It is the only tool to quantify the metabolite of tissue and the biochemical environment, so it is one of the major testing tools biomedical research in the 21st century. The parallel MRSI applies multiple coil array for parallel data acquisition, which reduce the acquisition time by reduce the phase encoding steps in one or two directions in space. The metabolite map reconstruction includes the two process of data reconstruction in k-space and spectrum quantification. We make research on them and give an improved SENSE reconstruction algorithm and a modified singular value decomposition method for spectrum quantification. First, we make research on the k-space data reconstruction algorithm. Because the sample density is sparse, aliasing artifacts will be seen in the reduced FOV in the reconstructed image if FFT method used directly. SENSE reconstruction algorithms can unfold the superposed voxels and combine into a full FOV image by the calculation of the coil sensitivity with the body coil reference scan data. SENSE-SI is studied carefully and improvement on the SENSE reconstruction algorithm is get. The mask with the same size as the object is added in the calculation of the distinct spatial sensitivities of each coil, which can improve SNR by using mask to extract the useless voxel and can improve the speed of reconstruction deeply. The accuracy and the superiority on SNR of the improved SENSE reconstruction algorithm are verified under MATLAB environment. A 3D cylinder phantom model is created to simulate 3D MRI using 2D SENSE with the total accelerate factor of four. The influence on SNR in the present of mask in x-y plan using different reduce factor is analyzed, with the results as following: 1) A higher SNR is seen in the case of using SENSE in x ? y slice than in y ? z slice. 2) If the two aliasing directions occur in the same plan as the mask, there is no significant difference when aliasing in two directions than in one direction. Else, higher the accelerate factor in one direction within the slice as the mask is, higher the SNR is. Then, the methods of spectrum quantification analysis in time-domain and frequency-domain are discussed. The former methods are based on the character that the MR signal can be expressed with the exponential model. The FID signal can be rearranged into a data matrix, and the kinds of the metabolites can be acquired by singular value decomposition (SVD). The ordinary SVD must decompose all the singular value, with the big cost of the time. Because the kind of metabolites is very limit, we put up a LSVD method to calculate the less singular to improve the rate according to the character of the Hankel matrix based on the Lanczos algorithm. When the SNR is higher than 10, the compution time is decreased by 13 times of the ordinary method. But the time is only half of the ordinary one when the SNR is lower than 1. The latter methods change the FID signal to frequency signal after the preprocessing such as the phase-correct. Peaks of different metabolites are picked applying the experiential information and the area of peaks is integrated to show the metabolite map. Finally, the software of MRSI reconstruction and spectrum quantification is developed under IDL 6.0 version, which can realize the processing of the raw data in clinic with the style of phase-encoding and SENSE. Numerical phantom model, k-space trajectories can be gotten from the software. And we can compare different methods of calculating coil sensitivity. It is helpful to study and research on new data reconstruction algorithms, and spectrum quantification analysis methods. The data in clinic are processed using the software, with the following results: the resolution of spectrum and metabolites are the same between the traditional phase-encoded SI and SENSE-SI, except the lower peak value and SNR in the latter. The NAA value in the tumor tissue is lower than the normal tissue, and the LAC peak can be seen in the abnormal area. In a word, the parallel MRSI technologies reduce the scan time at preserved spectral and spatial resolution, without the defects of the other fast MRSI methods. These improvements in data acquisition and image reconstruction provide a potential value of metabolic imaging as a clinical tool.