基于拉曼光谱技术结合机器学习算法分析创伤后应激障碍

本文采用血清拉曼光谱方法探究了创伤后应激障 碍(PTSD)的发病率与创伤环境中暴露时间 长短相关联的原因与机理分析。在本实验中以大鼠为研究对象,测试了28例延长创伤环境暴露时间 和27例非延长创伤环境暴露时间的大鼠血清样品。在被测大鼠血清拉曼光谱中特征峰的暂定 分配表明各组 之间的特定生物分子具有一定的差异性。采用主成分分析提取特征,用以减少高维光谱的维 数,来提高判 别模型的速率,并对提取后的特征分别采用BPNN,ELM和SVM进行分类,其中SVM算法又分别用 线性 核函数、多项式核函数和RBF核函数(高斯核函数)三种不同的核函数来建立模式识别模型 。PCA-RBF-SVM取得的结果最好,准确率为94.5%。研究结果表明, 不同创伤环境暴露时长的血清 拉曼光谱存在明显的差异性,暴露时长使其化学或分子物质发生了改变,继而可能导致创伤 后应激障碍的发病率发生变化。

A method based on serum Raman spectroscopy combined with SVM to explore the mechanism of PTSD

In this paper,serum Raman spectroscopy was used to investi gate the causes and mechanisms of the association between the incidence of post-traumatic stress di sorder (PTSD) and the length of exposure time in the trauma environment.In this experiment, 28rat serum samples were tested with prolonged traumatic environment exposure time and 27rat serum samples for non-prolonged traumatic environment exposure time.The tentative assignments of the characteristic peaks in the serum Raman spectra of the tested rats indicate that certain biomol ecules between the groups have certain differences.Principal component analysis is used to extract features to reduce the dimension of high-dimensional spectra to enhance the rate of discriminant model s.The extracted features were classified with BPNN,ELM and SVM.The SVM algorithm established p attern recognition model with the Linear kernel,Polynomial kernel and RBF kernel respectively.Th e PCA-RBF-SVM obtained thebest results with an accuracy rateof 94.5%.Theresult s of this study indicate that there is a significant difference in the serum Raman spectra with different exposure time in traumatic environments.The exposure time can change the chemical or mol ecular substances,whichmay leadto changes in theincidenceof PTSD.