The Research Of The Magnetic Resonance Imaging Reconstruction Based On Compressed Sensing

Magnetic resonance imaging as an extremely important imaging technique in themedical field, has advantages of lesion localization precision and high security for human.But the long data acquisition time become the bottleneck of wide application. It is urgent tofind a fast imaging methods, in the premise of ensuring the quality of imaging. Compressedsensing is a new signal processing method which can abandon the redundant information inthe current information sampling. Under the conditions of sparse or compressible, samplingsignal can accurately reconstruct original signals form a small quantity of measurements.So it improves the speed of sampling. This provides a new way to solve the key issues inmagnetic resonance imaging. Therefore, we apply the compressed sensing to the magneticresonance image reconstruction, in the foundation of the deeper research on many theories.And the major work and innovations in this paper include the following aspects:(1) In order to solve the image’s aliasing artifacts caused by the traditionalsub-sampling, it is proposed a way named random variable density sampling. Through thecomparison of several commonly used sampling trajectory, we adopt a kind of mode nameda variable density Cartesian random sampling mode, then we instruction this mode. Theresults show that the variable density Cartesian random sampling mode achieved goodresults based on compressed sensing MRI.(2) In order to solve problems of traditional two-dimensional wavelet transform cannot sparsely represent curves and edges. In this paper, we introduce a geometric imagetransform, the shearlet, to overcome this shortage. after discussing the implementation ofthe, reconstruction accuracy and sparsity of the discrete shearlet transform are analyzed.Experimental results shows that, discrete shearlet transform in the retention coefficient of1%, the value of SSIM is better than wavelet transform on average0.06at the samesampling rate. At the same time, discrete shearlet transform can improve quality ofreconstructed image and preserve more information about texture and edge.(3) signal reconstruction algorithms based on compressed sensing are high complexity,can not conducive to real-time processing. In view of the above problems, we propose adesign for interior point method that is based on the Field Programmable Gate Array(FPGA)hardware platform. We improvements the primal-dual corrector interior method,make it more suitable for FPGA of parallel processing. The array structure of the conjugategradients algorithm completes the main operations, and the parallel and pipelinecoprocessor effectively take advantage of the inherent parallelism of the algorithm and theparallel structure of FPGA. Thus, it can speedup about21times over the software versionof the same algorithm. (4) Sparse MRI image reconstruction algorithm contains a large number of floatingpoint arithmetic based on Compressed Sensing theory. It can be more time-consuming thantraditional inverse Fourier reconstruction. To solve this problem, the existing sparse MRIreconstruction algorithm is made practical by parallelizing it under the framework ofNVIDIA CUDA using GPU. The experimental results show that, based on the GPU, thisplatform provides very fast iterative reconstruction. Reconstruction of10242MRI canobtain speedups of24and49in the double precision floating point calculation and sigleprecision floating point calculation.