Savitzky-Golay平滑结合多元散射校正应用于甲醛固定的深静脉血栓红外光谱数据预处理  

Application of Savitzky-Golay Smoothing Combined with Multiple Scattering Correction in Infrared Spectral Data Preprocessing of Deep Venous Thrombosis Fixed with Formaldehyde

作　　者：刘琪 杜宇[1] 蒋金芳 陈绍璞 胡圣祥 谢嘉珑 Liu Qi;Du Yu;Jiang Jinfang;Chen Shaopu;Hu Shengxiang;Xie Jialong(Department of Forensic Pathology,Criminal Investigation Police University of China,Shenyang 110854,China;Shenzhen City,Yantian District Public Security Bureau,Shenzhen 518000,China)

机构地区：[1]中国刑事警察学院法医病理教研室,辽宁沈阳110854 [2]深圳市公安局盐田分局,广东深圳518000

出　　处：《山东化工》2024年第11期126-130,共5页Shandong Chemical Industry

基　　金：公安部技术研究计划项目(2022JSYJC25,2017JSYJC06,2014 JSYJB13);辽宁省自然科学基金项目(2021-MS-143);中国刑事警察学院研究生创新能力提升项目(2022YCZD03)。

摘　　要：为探究甲醛固定血栓样品红外光谱数据的最优预处理方案,应用傅里叶变换红外光谱(FTIR)技术,采集经甲醛固定的大鼠深静脉血栓组织红外光谱数据,选择偏最小二乘回归法建立红外光谱分析机器学习模型,测试Savitzky-Golay(S-G)平滑、基线校正、多元散射校正、标准正态变化、矢量标准化等常用的光谱预处理方式对光谱数据的处理效能。结果显示,大鼠深静脉血栓组织红外光谱的最优预处理方案为先采用S-G平滑,再进行多元散射校正,其中S-G平滑最佳参数组合是阶数为2、项数为3、点数为17。预处理后预测决定系数(R_(p)^(2))为0.948,预测均方根误差(RMSEP)为0.486 d,模型预测能力较原始光谱显著提升。In order to investigate the optimal preprocessing scheme for the infrared spectral data of formaldehyde-fixed thrombus samples,Fourier transform infrared(FTIR)was applied to collect the infrared spectral data of formaldehyde-fixed rat deep venous thrombosis tissues,and a machine learning model for infrared spectroscopy was established by selecting the partial least square regression method to test the efficacy of commonly used spectral preprocessing methods for processing spectral data,such as Savitzky-Golay(S-G)smoothing,baseline correction,multiple scattering correction,standard normal variation,vector normalization,and so on.,standard normal variate,vector normalization,and other commonly used spectral preprocessing methods on spectral data.The results showed that the optimal preprocessing scheme for the infrared spectra of rat deep venous thrombosis tissues was S-G smoothing followed by multiple scattering correction,in which the optimal parameter combinations of S-G smoothing were derivative order 2,polynomial order 3,and smoothing point 17,and the coefficient of determination of the prediction(R_(P)^(2))after preprocessing was 0.948,and the root mean squared error of prediction(RMSEP)was 0.486 d,which significantly increased the predictive ability of the model compared with that of the original spectra.

关 键 词：傅里叶变换红外光谱 Savitzky-Golay平滑 多元散射校正 深静脉血栓 甲醛固定 

分 类 号：O657.3[理学—分析化学]