Zhang Yuanzhi, Hou Huayi, Zhang Yang, Wang Yikun, Zhu Ling, Dong Meili, Liu Yong
Institute of Applied Technology, Hefei Institutes of Physical Science, Chinese Academy of Sciences, Anhui Provincial Engineering Technology Research Center for Biomedical Optical Instrument, Hefei, Anhui 230088, China.
Wanjiang Center for Development of Emerging Industrial Technology, Tongling, Anhui 244000, China.
Biomed Opt Express. 2018 Mar 22;9(4):1795-1808. doi: 10.1364/BOE.9.001795. eCollection 2018 Apr 1.
In order to reduce the influence of scattering and absorption on tissue fluorescence spectra, after tissue fluorescence and diffuse reflectance in different tissue optical properties were simulated by the Monte Carlo method, a tissue intrinsic fluorescence recovering algorithm making use of diffuse reflectance spectrum was developed. The empirical parameters in the tissue intrinsic fluorescence recovering algorithm were coded as a particle in the solution domain, the classification performance was defined as the fitness, and then a particle swarm optimization (PSO) algorithm was established for empirical parameters optimization. The skin autofluorescence and diffuse reflectance spectra of 327 subjects were collected in Anhui Provincial Hospital. The skin intrinsic autofluorescence spectra were recovered by using the empirical approach and the integration area of the spectra were calculated as fluorescence intensity. Receiver operating characteristic (ROC) analysis for fluorescence intensity was applied to evaluate the classification performance in type 2 diabetes screening. In addition, a support vector machine (SVM) method was implemented to improve the performance of the classification. The results showed that the sensitivity and specificity were 32% and 76% respectively, and the area under the curve was 0.54 before recovering, while the sensitivity and specificity were 72% and 86% respectively, and the area under the curve was 0.86 after recovering. Furthermore, the sensitivity and specificity increased to 83% and 86% respectively when using linear SVM while 84% and 88%, respectively, when using nonlinear SVM. The results indicate that using the tissue fluorescence spectrum recovery algorithm based on PSO can improve the application of tissue fluorescence spectroscopy effectively.
为了降低散射和吸收对组织荧光光谱的影响，采用蒙特卡罗方法模拟了不同组织光学特性下的组织荧光和漫反射，在此基础上开发了一种利用漫反射光谱的组织固有荧光恢复算法。将组织固有荧光恢复算法中的经验参数编码为解域中的粒子，将分类性能定义为适应度，进而建立粒子群优化（PSO）算法对经验参数进行优化。收集了安徽省立医院327名受试者的皮肤自发荧光和漫反射光谱。采用经验方法恢复皮肤固有自发荧光光谱，并计算光谱积分面积作为荧光强度。应用荧光强度的受试者工作特征（ROC）分析来评估2型糖尿病筛查中的分类性能。此外，还采用支持向量机（SVM）方法提高分类性能。结果显示，恢复前灵敏度和特异度分别为32%和76%，曲线下面积为0.54；恢复后灵敏度和特异度分别为72%和86%，曲线下面积为0.86。此外，使用线性SVM时，灵敏度和特异度分别提高到83%和86%；使用非线性SVM时，灵敏度和特异度分别为84%和88%。结果表明，基于粒子群优化的组织荧光光谱恢复算法能有效提高组织荧光光谱的应用效果。
