A Study Of Non-Cartesian MRI Algorithms Based On Artificial Sparsity

Magnetic resonance imaging(MRI)plays an increasingly important role in modern medical imaging.Imaging speed,however,has become the major limitation of MRI which results in less practical clinical applications,including the imaging for some patients in dysphoria.Furthermore,slow imaging speed also impairs the imaging quality of dynamic MRI,in which there is a trade-off between spatial and temporal resolutions.Therefore,it is of great meaningful to accelerate the imaging speed.To accelerate the MRI procedure,a variety of fast imaging techniques have been proposed.Among these ideas,a commonly-used acceleration technique acquires partial k-space data instead of full sampling.A number of approaches have previously been proposed,including partial Fourier imaging exploiting the property of conjugate symmetry in k-space,non-Cartesian imaging which samples along a non-Cartesian trajectory,parallel imaging(PI)considering the redundant information among multiple coil elements,and compressed sensing MRI that employs the sparsity of images.Artificial sparsity is a sparsity stragedy based on the property of PI having an improved performance when the to-be-reconstructed image is sparse.Auto-calibrated PI can inherently consider the coil sensitivity and image content to improve the quality of reconstruction.Artificial sparsity has been widely used in Cartesian MRI,but,in non-Cartesian MRI,little progress has been achieved.Due to the less sensitive to motion,more uniformly-distributed aliasing artifacts and higher signal-to-noise(SNR)efficiency,non-Cartesian has attracted much attention from researchers.Therefore,it is of great significance to extend artificial sparsity to non-Cartesian imaging.In this work,taking radial sampling as an example,a series of artificial sparsity algorithms have been proposed,as the following:(1)Artificial sparsity algorithm for static non-Cartesian imagingThree artificial sparsity schemes were designed for linear-angle radial imaging and golden-angle radial imaging,among which the optimal solution was selected.After extracting artificial sparsity data,an artificial-sparsity-based algorithm named ARTS-GROWL was proposed.Afterwards,it was verified using simulated brain data and in vivo data.The experimental results demonstrated that artificial sparsity improved the image quality of non-Cartesian parallel imaging with SNR improved and normalized root-mean-square error decreased.(2)Artificial sparsity algorithm for dynamic non-Cartesian imaging In dynamic MRI,a three-dimensional algorithm based on dynamic artificial sparsity was proposed.Compared to conventional parallel imaging,this algorithm achieved better reconstruction.Compared with two-dimensional artificial sparsity,this technique generated reconstructed images with less artifacts and lower noise.Compared to new k-t algorithms(iGRASP,k-t SPIRiT),the proposed scheme resulted in better or comparable image quality with lower computational cost.This may help to improve the clinical applicability of dynamic MRI.(3)Motion-sorted dynamic artificial sparsity algorithmBy leading in motion information sorting technique,previous dynamic artificial sparsity had been optimized.This algorithm was verified in liver dynamic contrast-enhanced data with golden-angle radial sampling scheme.The experimental results demonstrated that better reconstructed results can be obtained after incorporating motion-sorted technique into original dynamic artificial sparsity scheme.The motion artifacts were reduced,the SNR was enhanced,and the noise in the background was suppressed.It is demonstrated that parallel imaging results can be improved in static or dynamic non-Cartesian MRI by leading in artificial sparsity,and image quality can be further enhanced after motion sorting.