基于随机变量交替方向乘子法的荧光分子断层成像

发布时间：2018-06-16 13:24

本文选题：医用光学 + 荧光分子断层成像　；参考：《光学学报》2017年07期

【摘要】：增加测量信息可以有效降低荧光分子断层成像(FMT)重建的病态性,但随着数据增多,重建耗时也会显著增加。为了降低FMT重建的病态性和提升大规模数据集下的重建效率,结合对偶坐标下降法(DCA)和交替方向乘子法(ADMM)提出了一种改进的随机变量的交替方向乘子法重建优化方法。在原始ADMM方法的基础上,增加了一个随机更新规则,在每次迭代中只需要一个或者几个样本,就可加速收敛,使目标函数快速得到最优解,从而达到快速重建的效果。设计了数字鼠仿真实验和真实鼠实验,实验结果表明,所提方法在保证FMT重建图像精度的同时,显著提高了重建效率。
[Abstract]:Increasing the measurement information can effectively reduce the pathological feature of FMT reconstruction, but with the increase of data, the reconstruction time will increase significantly. In order to reduce the ill-condition of FMT reconstruction and improve the efficiency of reconstruction in large scale data sets, an improved method for reconstruction optimization of random variables by alternating direction multiplier method (ADMMMMM) and dual coordinate descent method (DCA) is proposed. On the basis of the original ADMM method, a random update rule is added. Only one or several samples are needed in each iteration, which can accelerate the convergence and make the objective function obtain the optimal solution quickly, so as to achieve the effect of fast reconstruction. Digital rat simulation experiments and real mouse experiments are designed. The experimental results show that the proposed method not only ensures the accuracy of FMT image reconstruction, but also improves the reconstruction efficiency significantly.
【作者单位】：西北大学信息科学与技术学院;
【基金】：国家自然科学基金(61640418,61601363)
【分类号】：TP391.41

【参考文献】