Rapid Non-Cartesian Reconstruction Algorithm For Magnetic Resonance Imaging

Magnetic Resonance Imaging (MRI) has been widely used in clinical practice due to its non-radiation, high resolution, multi-orientation, multi-parameter, etc.. Compared to Cartesian sampling, non-Cartesian sampling offers some potential advantages, such as fast acquisition and insensitivity to motion and flow. The advantage is more obvious in the dynamic cardiac imaging and function brain imaging. In addition, many techniques about motion correction based on non-Cartesian sampling take full advantage of navigation information integrated into its acquisition, which can eliminate motion artifacts in a very effective way.However, a major reason that constraints of non-Cartesian sampling widely applied in the daily clinical practice is the low speed of image reconstruction. Because the sample points do not fall onto the Cartesian grid points, Fast Fourier transform algorithm can not be used directly. So some more advanced reconstruction algorithms must be used in order to reconstruct optimal image. But these algorithms are very complicated and calculation intensive.This paper focused on the PROPELLER trajectory with other non-Cartesian sampling trajectory. In order to promote their wider application in clinical use, Several methods of non-Cartesian reconstruction to improve the speed of the algorithm were studied.The main research achievements include:(1) The effect of different density compensation on image quality is analyzed. The conclusion is drawn that Jackson density compensation function is a better choice in gridding reconstruction for PROPELLER after considering calculating time and image quality.(2) The method of rapid calculation of density compensation function is developed based on PROPELLER and Radial trajectory, which is roughly400times faster than the conventional method. This method enables real-time calculation of density compensation function, and fortunately the image quality is not significantly affected.(3) NUFFT (nonuniform fast Fourier transformation) is useed in each iteration step, so the iterative reconstruction algorithm is greatly accelerated.(4) The iterative method is applied in reconstruction for PROPELLER trajectory. The experimental results show that the iterative method can reduce ring artifacts, and improve the ratio of signal and noise and homogeneity in reconstructed image. Accelerated using NUFFT, the speed of the method is improved more than40times.(5) A new gridding algorithm is designed based on CUDA for PROPELLER trajectory. In the algorithm, a new data structure for storing sampling data is used. And a new search method in convolution interpolation step is proposed. Averagely gridding algorithm on GPU is8times faster than on CPU, but average root mean square error is lower than3x10-4between two images reconstructed using CPU and CPU algorithm.(6) Iterative algorithm is implemented on GPU based on discrete-discrete MRI imaging model.By reconstructing Spiral data the results showed that, after GPU acceleration, the computational speed of iterative reconstruction has been improved70times compared to the algorithm on CPU. But normalized root mean square error is lower than5×10-4In conclusion, many methods are proposed to improve the speed of non-Cartesian reconstruction, so it will help a lot to promote their application in clinical practice. These methods took full use of MRI imaging theory and feature of non-Cartesian sampling trajectory, and combined software technique with hardware echnique.