The Study Of Fast Dynamic Magnetic Resonance Imaging Based On Compressed Sensing Technique

Limited by the hardware performance and human physiological limitations, the scan timeof Magnetic Resonance Imaging (MRI) is relatively longer, especially in dynamic MRIapplications. Compressed Sensing (CS) theory, breaking down the Nyquist Criterion, wasproposed by Donoho et al in2004. It could reconstuct the signal for the highly undersampleddata by solving the sparse regularization problem. Dynamic MRI applications usually havemany redundancies along temporal direction, and thus providing the possibility for applyingcompressed sensing technique. This study is mainly focused on the problems of k-t FOCUSS(k-t space FOCal Underdetermined System Solver) algorithm and its improvement and fastreconstruction for3D data based on ROF (Rudin-Osher-Fatemi) model:(1) Study the theory of k-t FOCUSS and its application in dynamic MRI. Model-basedreconstruction method called k-t BLAST/SENSE provides a better spatio-temporal resolutionin dynamic MRI. k-t FOCUSS combines strategies from compressed sensing and from k-tBLSAT/SENSE to improve the quality of reconstruction. There are two main parts in k-tFOCUSS: image prediction and residual encoding. Averaged time frame was used as theprediction frame. Residual encoding was obtained by substracting the zero-paddingreconstruction from prediction frame. Experiment results show that the quality of k-t FOCUSSreconstruction is much better than k-t BLAST/SENSE.(2) Study the possibility of applying Motion Estimation and Motion Compensation(ME/MC) technique in k-t FOCUSS for dynamic MRI. A new ME/MC method with OBMC(Overlapped Block Mmotion Estimation) algorithm was improsed to improve the motioncompensation. The simulation result of dynamic cardiac data reconstruction demonstrates thatthe prediction by ME/MC without OBMC will cause terrible block-like artifacts, which can beeffectively suppressed by ME/MC with OBMC method.(3) Study the fast compressed sensing reconstruction method for3D dynamic data. TheROF model was combined with CS technique. The FBS (Forward–Backward Splitting) andAFBS (Accelerated Forward–Backward Splitting) algorithm was studied and applied in fast3Dreconstruction. The3D ROF model utilizes the redundancy among undersampled3D data tointerpolate the un-sampled data. The3D ROF model expoits the redundancy between farmes bycalculating the gradient sparse, FBS and AFBS algorithm were used to reolve the3D ROFmodel. There are two main steps of FBS algorithm, that is, data updating and median filtering.AFBS is the optimization of FBS and can accelerate the conbergence. Experiment result shows that compressed sensing with3D ROF model can reduce the image artifact. The quality ofAFBS reconstruction is similar to that of FBS reconstruction, but with faster convergence.