| Magnetic resonance imaging (MRI) is a non-invasive detection of no damageimaging techniques. Unlike traditional imaging techniques such as CT, MRI has noionizing radiation, and at the same time, it can obtain images of various parameterswith high contrast, etc. With these advantages, MRI becomes an important clinicalmedicine diagnostic tool. Nevertheless, it has its own drawbacks that the scan time isa litter long and the imaging speed a bit slow, which limit its application in manyclinical situations. Thus, since the birth of MRI technology, people have beenexploring the fast MRI methods. Although, there has developed a number of methods,such as spiral acquisition, radial acquisition, parallel imaging methods, and thesemethods in some extent, greatly improving the speed of imaging, but still can notmeet requirements for high clinical situations.In the year of2006, D Donoho et al proposed the Compressed Sensing theory(CS), which is different from the traditional sampling theorem. According to the CS,the original signal can be recovered from a small amount of sampling data. Therefore,the magnetic resonance imaging technology based on compressed sensing was thenproposed. Now, CS has a lot of the new algorithm on MRI, but these algorithms are inthe simulation stage. Apply it in the MRI sans there are two aspects to be considered.One is the sequence design; the second is to achieve CS reconstruction algorithmproject realization. Our approach is to solve the problem that is the second aspect.In this paper, we studied the principle of magnetic resonance imaging at first.Then explored the conditions it needs to satisfy the theoretical framework ofcompressed sensing and combined it with magnetic resonance imaging methods.While we designed the measurement matrix that is the sampling trajectories, theanalysis of non-coherence that required in the design of samplings and a variety ofMRI images sparse transforms were studied. In the sampling, we designed severalsampling trajectory, but because of hardware limitations and requirements of CS,two-dimensional MR images, generally randomly collected using variable density1D;And in sparse transformation, the simple image transformation using Total Variationthat are able to meet the requirements, but other complex images you need totransform sparse wavelet transform. Of course, the compressed sensing reconstructionalgorithm was also in-depth analyzed.Then, the implementation method of CS-MRI using the C++language wasstudied, and several key issues in the implementation were analyzed. Meanwhile, anumber of important functions are described in details, such as the function of generating model data, fast Fourier transform, the sampling trajectory functions, thetotal variation difference operation, wavelet sparse transform and the other corefunction in the implementation of CS algorithms.Finally, a reconstruction software of CS_MRI that was based on dialog wasdeveloped using MFC framework in VC++6.0platform. On the software, we canestablish phantoms with different resolutions or import the data collected by MRscanner. If the data were created by us, we should change them into k-space data withFFT operations, and saved them in the appropriate file. And we can select samplingtrajectories that can simulate compressed sensing sparse collection and we can selectthe image sparse transform methods. There are two kinds of reconstruction methodsthat we can choose. At last, we have tested the CS-MRI software; images can besuccessfully reconstructed with high quality by using the algorithm. The result showsthat the engineered implementation of CS-MRI algorithm can accelerate the processto apply it on the scanner. It provides a reference for the CS algorithm combiningwith other algorithms, such as parallel compressed sensing imaging. |
PDF Full Text Request