Regularization-based Parallel Magnetic Resonance Imaging Technique For Highly Accelerated Data Acquisition

The long scan time is a bottle neck for the development of magnetic resonanceimaging (MRI). The improvement of the gradient system can directly accelerate everysingle data acquisition and reduce the scan time. However, this method has nearlyreached its limits for scan time reduction, due to the physiological problem induced bythe magnetic gradient. Parallel imaging, on the other hand, shortens the scan time byreducing the amount of the acquired data. It does not rely on the magnetic gradientand has no physiological problems, which makes parallel imaging available for almostall MRI applications. Therefore, it is of great significance to research and develop thehighly accelerated parallel imaging technique for fast MRI.The di?culty of achieving highly accelerated parallel imaging lies in that its re-construction is an ill-posed inverse problem. For this reason, the reconstructed imagesu?ers from severely amplified noise when the acceleration is high. To solve the prob-lem, this thesis proposes two regularization methods. To maintain the computationale?ciency of Tikhonov regularization while reducing its blurring e?ect, we appliedthe wavelet-based multivariate linear regularization to parallel imaging reconstruction.Furthermore, we proposes a parameter estimation method based on power signal-noiseratio, which enables the automatic and adaptive parameter estimation of multivariateregularization. To achieve a high quality reconstructed image, we proposes the non-linear coherence regularization, which significantly outperforms current regularizationmethods in preserving image details. Besides, the coherence regularization can notonly be used to transform various image filtering methods into regularization, but alsobe used to design regularization directly based on image characteristics. Therefore, itprovides a new framework for designing regularization.The estimation of receiver coil sensitivity is also critical for parallel imaging re-construction. This thesis proposes a nonlinear method to improve the accuracy of coilsensitivity estimation. This method addresses the problems of resolution loss and Gibbs ringing in the standard linear method, so that the spatial information of coil sensitivityis better preserved. With this sensitivity information, the parallel imaging reconstruc-tion exhibits fewer error than the standard method.There exists an intrinsic trade-o? between image quality and imaging speed inMRI. Under some situations, parallel imaging technique can not only boost the imag-ing speed, but also improve the image quality without obviously sacrificing the imag-ing speed. To demonstrate the versatility of the proposed method, we applied it to echoplanar imaging, cardiac imaging and temperature mapping. The experimental resultsdemonstrated that the proposed method could e?ciently improve the spatial or tem-poral resolution of MR imaging, which would potentially be available in a variety ofapplications.