摘要
为提高荧光分子断层成像(FMT)的重建精度,将非凸LP(0<P<1)正则化引入到FMT重建中。为有效求解包含非凸惩罚项的优化问题,结合加权同伦算法(WHA)提出了快速迭代重建算法,将LP正则化问题转化为一系列加权的L1正则化问题求解。异质仿体上的仿真实验表明,LP(0<P<1)相较L1正则化重建在荧光目标定位和定量方面都有提高。而且通过比较6种不同的正则子,评估了参数P的选择对重建结果的影响,结果表明当1/3≤P≤1/2时,非凸正则子算法可以取得最优的重建结果。
To improve the accuracy of fluorescence molecular tomography（FMT）, non-convex LP（0 〈P 〈1）regularization is introduced into 3D reconstruction. By transforming the non-convex optimization problem into a series of weighted L1 regularization problems, an iterative reconstruction scheme based on the weighted homotopy algorithm（WHA） is proposed to efficiently solve the optimization problem with non-convex LP（0〈 P 〈1）penalty. Simulations on a heterogeneous phantom demonstrate that the proposed LP（0〈 P〈 1） regularization algorithm has better performance than L1 regularization in terms of location error and fluorescent yield.Comparison results with six different regularizers assess the impact of parameter P on the reconstruction. The results demonstrate that the best reconstruction results can be obtained with 1/3 ≤ P ≤ 1/2.
出处
《光学学报》
EI
CAS
CSCD
北大核心
2015年第7期252-259,共8页
Acta Optica Sinica
基金
国家自然科学基金(61372046
61401264)
关键词
医用光学
非凸正则化
加权同伦算法
荧光分子断层成像
重建算法
medical optics
non-convex regularization
weighted homotopy algorithm
fluorescence molecular tomography
reconstruction algorithms
