针对荧光扩散层析成像中荧光光源的定位误差较大和形态学信息不完整的问题,提出一种基于组稀疏正则化的同时代数重建技术(GSR-SART)算法。该算法利用图像的非局部自相似性和局部稀疏性构造自适应相似组;然后,将相似组作为基本单元,学习自适应字典;最后,采用迭代收缩阈值算法求解目标函数。实验结果表明,所提算法在峰值信噪比和方均根误差的结果上比其他先进算法有较大的提升。

A simultaneous algebraic reconstruction technique based on group sparse regularization (GSR-SART) algorithm is proposed in this study to address the problems of large positioning error of the fluorescent light source and incomplete morphological information in fluorescence diffuse optical tomography. The algorithm uses nonlocal self-similarity and intrinsic local sparsity to construct the self-adaptive similar group. Then, the similar group is considered the basic unit to learn the adaptive dictionary. Finally, the target function is solved using the iteration shrinkage threshold algorithm. The experimental results show that compared with the other advanced algorithm, the proposed algorithm yields better results in terms of peak signal-to-noise ratio and root mean square error.