Research On Parallel MRI Reconstruction Based On Compressed Sensing And SENSE

Parallel magnetic resonance imaging is a method which is using multiple receiver coil to acquire signals to reduce the phase encoding times, thereby reducing signal acquisition time and speed up the imaging. Wherein, SENSE is a method which is currently more mature and widely used in parallel MRI. Compressed sensing is a new theory in the field of signal processing, it can take advantage of the sparse nature of the signal to reconstruct the original signal. If the signal is sampled randomly, original signal can be perfectly reconstructed even if sampling rate lower than the Nyquist frequency. This paper focuses on Compressed Sensing and SENSE studies and evaluate their reconstruction effect. In addition, the combination of two technologies can greatly improve the sampling rate. At the same time, the combination can also solve the problem which parallel magnetic resonance imaging in higher speed ratio the SNR fell sharply. In this paper, the main work includes:(1) Based on sparsity constraint, combined the compressed sensing and sensitivity encoding parallel MRI reconstruction and detailed analysis the random sampling scheme of this framework, sensitivity reconstruction method and the accelerate ratio and so on. Experimental results show that this framework has good performance when in a higher accelerate ratio.(2) Due to the above framework's compressed sensing stage having a TV and LI complex regularization constraint, solving it by traditional conjugate gradient type algorithms require a lot of computing time. So I applying fast composite splitting algorithm in the above framework to split solving the multiple regularization terms. Experimental results show this algorithm can accelerate the reconstruction, at the same time, can achieve a better effect in reconstruction of the 1-D k-space random sampling acquisition.(3) Traditional sparse MRI reconstruction methods using wavelet transform as L1 regularization item. Caused by a shortcoming of wavelet in multiscale geometric analysis, I take the Contourlet as a new regularization item introduced into the reconstruction framework and implement by the modified FCSA algorithm. In addition, applying it to the compressed sensing parallel MRI framework. Experimental results show good performance have achieved.