具有噪声鲁棒性的三维磁性纳米粒子成像快速重构方法

Novel Noise-Robust and Fast Reconstruction Method For 3D Magnetic Particle Imaging

为提升磁性纳米粒子三维成像和重构速度,降低三维重构对采样投影数据完备性要求,针对含噪声投影数据三维重构优化问题,本文提出了一种具有噪声鲁棒性的三维磁性纳米粒子成像快速重构方法(noise-robust 3D sparse sampling magnetic particle imaging,3D NRSS-MPI).该算法通过求解一个由MPI投影成像正向模型l2范数和稀疏正则约束建立的凸优化问题实现3D MPI图像重构.模型不受MPI扫描轨迹限制,为不断发展的MPI新技术提供了普适性的基础模型;建立的三维全变分稀疏算子(3D total variation sparse operator, 3D TV Sparse Operator)利用MPI先验信息提升含噪MPI投影数据三维重构的鲁棒性,且可以进行无矩阵运算,大幅提升了运算效率.通过点源和冠状血管模型成像实验表明,在1/4欠采样下,本文3D NRSS-MPI方法可以有效消除重构图像星状伪影,获取较高的图像信噪比,同时冠状动脉重建结构相似性超过0.701,可以准确地对欠采样、有噪声的MPI数据进行快速而稳健的重建,有效缩短了4倍成像和重构时间.

To improve the 3D imaging and reconstruction speed of Magnetic Particle Imaging(MPI), reduce the requirement of 3D refactoring to the completeness of sampled projection data, a novel Noise-Robust 3D Sparse Sampling Magnetic Particle Imaging(3D NRSS-MPI) method was proposed. The algorithm was arranged to reconstruct 3D MPI noisy data by solving a convex optimization problem formed with the l2 norm and sparse regular constraint of MPI projection imaging. Eliminating the limit of MPI scanning trajectory, the proposed method was designed as a universal basic model for the developing MPI technique. Taking the advantage of MPI priori information to improve the 3D reconstruction robustness of noisy MPI projection data, 3D total variation sparse operator was established to realize matrix-free operation, improving the efficiency of operation. The results of point source and coronary phantom imaging experiments show that the proposed 3D NRSS-MPI method can effectively eliminate the reconstructed image star artifacts at 1/4 undersampling, obtain a higher image signal-tonoise ratio, and make the coronary reconstruction structure similarity exceed 0.701, which can accurately reconstruct the undersampled and noisy MPI data, effectively shortening the imaging and reconstruction time by 4 times.