Fast and robust three-dimensional fluorescence source reconstruction based on separable approximation and adaptive regularization
Fast and robust three-dimensional fluorescence source reconstruction based on separable approximation and adaptive regularization
复制标题
DOI：
10.1016/j.optcom.2012.08.047
复制
发表时间：
2012-11
影响因子：
2.4
通讯作者：
Z. Xue;C. Qin;Qian Zhang;Xibo Ma;Xin Yang;Jie Tian
中科院分区：
文献类型：
--
作者：
Z. Xue;C. Qin;Qian Zhang;Xibo Ma;Xin Yang;Jie Tian
本文提出了一种基于可分离逼近和自适应正则化的快速、鲁棒的荧光分子层析重建方法。通过可分离逼近可以有效地建立和求解子问题，自适应正则化也可以加速收敛过程。众所周知，正则化参数对结果有重要影响，自动找到最优或接近最优的正则化参数是一项具有挑战性的任务。为了解决这个问题，在所提出的方法中的正则化参数更新的启发式，而不是手动或经验确定。该方法的自适应正则化策略几乎不需要考虑正则化参数的选择，就可以实现精确的重建。相比之下，不适当的正则化参数的选择可能会导致较大的定位误差的三种对比方法。该方法具有鲁棒性强、对参数不敏感等优点，可提高重建精度.此外，所提出的方法是约1-2个数量级的速度比常用的荧光层析重建的对比方法。此外，还研究了不同初始未知值和不同噪声水平下的可靠性。最后，通过小鼠模型的物理实验进一步验证了该方法的实际应用潜力。
In this study, a fast and robust reconstruction method based on the separable approximation and the adaptive regularization is presented for fluorescence molecular tomography. The subproblems can be established and solved efficiently through separable approximation, and the convergence process can be also accelerated by adaptive regularization. As is well known, the regularization parameter has an important impact on the results, and finding the optimal or near-optimal regularization parameter automatically is an challenging task. To solve this problem, the regularization parameter in the proposed method is updated heuristically instead of being determined manually or empirically. This adaptive regularization strategy of the proposed method can perform accurate reconstruction almost without worrying about the choice of the regularization parameter. By contrast, improper choice of the regularization parameter may cause larger location errors for the three contrasting methods. The proposed method is proved robust and insensitive to parameters, which can improve the reconstruction accuracy. Moreover, the proposed method was about 1–2 orders of magnitude faster than the contrasting methods commonly used in fluorescence tomography reconstruction. Furthermore, reliable performance on different initial unknown values and different noise levels was also investigated. Finally, the potential of the proposed method in a practical application was further validated by the physical experiment with a mouse model.