基于SENSE和GRAPPA的并行磁共振图像重建算法

Parallel magnetic resonance imaging reconstruction algorithm based on generelized autocalibrating patially parallel acquisitions and sensitivity encoding

并行磁共振成像技术降低了采样时间,提高了成像速度和图像分辨率,但重建图像信噪比有所下降。为此提出了一种基于SENSE(Sensitivity Encoding)和GRAPPA(Gene Relized Autocalibrating Patially Parallel Acquisitions)的并行成像算法,降低由于减少采样行而造成的信噪比损失。在采集较少K空间中心自标定(ACS)行基础上,先用GRAPPA算法拟合出更多中心数据估计得到较为精确的线圈灵敏度,采用共轭梯度法进行图像重建,得到质量较好的重建图像,进而结合估计的线圈灵敏度进行交替迭代优化,计算出误差较小、分辨率较高的最终重建图像。采用了不同加速因子的人脑磁共振K空间欠采样数据以验证该算法的重建性能。仿真实验结果表明,该算法重建出的MR图像从视觉效果上和定量对比结果上都优于已有算法。尤其是在加速因子较大、采样行数较少时可以重建出质量更高的磁共振图像,具有更低的归一化均方误差和更高的信噪比(能提高22%)。新算法降低K空间采样行的同时,提高了并行磁共振重建图像信噪比并降低了噪声干扰。

Parallel magnetic resonance （MR） imaging technology reduces sampling time, improves imaging speed and image resolution, but reduces the signal- to- noise ratio （SNR） of reconstruction image. A parallel imaging algorithm based on the sensitivity encoding （SENSE） and generelized autocalibrating patially parallel acquisitions （GRAPPA） was proposed to reduce the loss of SNR caused by the reducing of sampling time. GRAPPA algorithm was used to fit more missing central data in central K-space with a small number of autocalibration signal lines in order to estimate and generate more accurate coil sensitivities. The better reconstruction images were obtained by conjugate-gradient. And then combined with estimated coil sensitivities, alternating iterative optimization was carded out to calculate the final reconstruction image with small error and high image resolution. The sampling data in the human brain MR K-space of different accelerated factors were applied to verify the reconstruction function of the algorithm. The simulation experiences showed that MR images reconstructed by the proposed algorithm were better than those reconstructed by the conventional algorithm in visual effects and quantitative eomparis0n results. When the accelerated factors were larger and the number of sampling lines was small, the MR images with higher quality were reconstructed, with lower normalized mean squared error and higher SNR which could be improved by 22%. The proposed algorithm can reduce the number of sampling lines in K-space, and improve the SNR of parallel reconstruction MR images, and lower the noise interference.