High Resolution Dynamic MRI Based On Partial Separability Model And Compressive Sensing

Magnetic resonance imaging (MRI), which is based on the principle of nuclearmagnetic resonance (NMR) and allows high quality images with the functional,structural and lesion’s information of an organism, is a state-of-the-art imagingmodality. Compared with other imaging technologies, MRI has been widely used inclinical applications and scientific research fields with many advantages, such asmultiple parameters imaging, unprecedented soft tissue contrast and free harmfulradiation. However, conventional MRI suffers from its slow imaging speed and thushas difficulty in achieving high spatial resolution and high temporal resolution,which has limitation on some special applications such as real-time imaging ofcardiovascular system and so on. To address this issue, after the systematic study ofthe existing fast MRI methods based on signal processing theory, we adopted thePartially Separable Functions(PSF) model based on compressive sensing(CS) indynamic imaging. The main works and contributions of this thesis can be describedas follows:Firstly, the PSF model and CS theory is investigated respectively to improvethe MRI speed on the condition that maintains the reconstruction quality. Based onthe advantages of PSF and CS, we combines the PSF model and CS methodstogether to further accelerate the MR imaging speed. Both simulation experimentand in-vivo experiment has been implemented to verified the proposed method. Theresults show that our proposed method can not only overcome the disadvantage thatthe PSF model would cost a long pre-data scan time for imaging, but also providebetter reconstruction.Secondly, based on the study of noisy-distribution of pseudo-random sampledMR images,we introduce an approach that compute local regularization parameteradaptively to the method of the parallel MRI (pMRI) based on compressive sensing.The proposed approach includes three steps. First, we should obtain the referenceimage. Second, noisy-distribution of the reference image is evaluated in spacedomain. Third, the local regularization parameter can be computed adaptively. Compared to the traditional single-parameter method, Experimental results showthat the proposed method can maintain the reconstruction quality.