Research Of Parallel Magnetic Resonance Imaging And Obtation Of Phase Information

Magnetic resonance imaging(MRI) has been one of the most important tools for clinical diagnosis and medical research owing to it is non-invasive, non-ionising and the unique quality of soft tissue contrast. However, long scan time handicaps more widely using of MRI. Parallel MRI(pMRI) technique can efficiently speed up MRI scan while maintaining field of view(FOV) and spatial resolution. Although pMRI has been successfully used in various clinical application, however, only low acceleration factors can be achieved in practice because the images reconstructed by pMRI suffer from serious aliasing artifacts and noise amplification at high acceleration factors. Therefore, the scan speed and image quality of pMRI technique need to be meliorated.Phase information of complex MRI signal can represent nonuniformity of main field and the susceptibility of tissue, thus it has broad applying prospect. However, phase is usually ignored in most existing MRI applications due to the low signal-noise-ration(SNR) in low field MRI and phase wrap. The development of superconductivity and p MRI techniques notably improves the SNR of phase, and draws an increasing attention to the phase information in the high field MRI system. In the pMRI, the phase received from each coil is the sum of coil phase, background phase from the main field, and the phase of protons. Consequently, optimal coil combination method is necessary to obtain the real phase of protons, as well as to remove the coil phase and background phase. Nonetheless, standard coil combination method is currently vacant to accurately obtain the phase of protons.In order to improve the imaging speed and image quality, this dissertation focuss on the magnitude and phase image reconstruction in pMRI. Especially, the determination of interpolation window, the reduction of aliasing artifacts and noise in the K-space based reconstructed generalized auto-calibrating partially parallel acquisition(GRAPPA), as well as the phase obtaining from the K-space data acquired by multi-channel coils, are researched. Specifically, the research contents in this dissertation include the following four aspects:(1) Robust GRAPPA reconstruction using sparse multi-kernel learning and support vector machine(SVM) has been proposed to solve the problem that existing GRAPPA reconstructions heavily depend on the choice of interpolation window and other reconstruction and sampling parameters. In the proposed method, linear, polynomial and Gaussian kernels are concurrently employed to map the K-space data onto a higher dimension feature space. Then, the interpolation coefficients are estimated by SVM that can attain excellent trade-off between the fitting error and generalization error. Experimental results on phantom and in vivo data indicate that the proposed method can automatically achieve an optimized compromise between noise suppression and residual artifacts for various sampling schemes. Compared with traditional GRAPPA and nonlinear GRAPPA, the proposed method is significantly less sensitive to the interpolation window and kernel parameters.(2) Correlations between adjacent K-space data explored in existing K-space based pMRI methods are summarized and classified into three categories: forward, backward and self- constraints. Combining such three constraints, a separate self-constraint GRAPPA reconstruction was developed, in which the interpolation of missing K-space data is resorted to solving a system of linear equations. The proposed method constructs the separate self-constraint by separately using the acquired and missing K-space data where the self-constraitn with acquired data is used to adaptively calibrate the self-constraint coefficients, which can reduce artifacts and noise aroused from the data inconsistency. In view of the long distance between the interpolation source and target data at hight accelerations, the self-constraint is weighted to avoid potential overfitting. Further, the weight is automatically determined through the fitting error of such constraints in the auto-calibration signal(ACS) region. Experimental results demonstrate that the proposed method can yield better image quality than existing K-space based pMRI methods, and has the potential to offer improved parallel MRI reconstruction.(3) Nuclear norm-regularized K-space based pMRI reconstruction has been developed to reduce the noise in image reconstructed by current pMRI at high accelerations. The interpolation in pMRI is translated into a linear inverse problem in the proposed method to introduce the regularization. Meanwhile, the K-space data through all coils are re-ordered as a low-rank matrix to explore the constraint of low-rank. Then, the non-convex rank function is replaced with nuclear norm, and the reconstruction of missing K-space data is converted to optimize an unconstrained programme of non-convex function. Finally, the optimal solution is obtained by alternating direction method of multipliers. Experimental results of brain data demonstrate that the proposed method can efficiently suppress noise amplification in the conventional K-space based pMRI reconstruction.(4) Optiaml phase obtaining method through ant-convolution in K-space has been proposed to circumvent the plight that magnitude weighed based phase obtaining methods are affected by the noise and aliasing artifacts in the pMRI reconstructed magnitude image at accelerated sceneries. The proposed method firstly employs sum-of-squares and phase aligning method to yield a complex reference coil image that is subsequently used to calculate the coil sensitivity and its Fourier transform. Then, the coil K-space combining weights are computed using the truncated frequency data of coil sensitivity and the acquired K-space data. Finally, combining the coil K-space data using the combining weights gives the K-space data of proton distribution, and then the phase and magnitude information can be obtained after the inverse Fourier transform. Experimental results show that the proposed method can alleviate the phase cancellation in coil combination, and gives less wapped phase against the magnitude weighed based methods.