血红蛋白定量分析的数据集划分及预处理方法研究  

Dataset partitioning and pretreatment method of hemoglobin quantitative analysis

作　　者：张朱珊莹 蒙泳吉 曹汇敏[1] 张莉[1] 郑冬云[1] 谢勤岚[1] ZHANG Zhushanying;MENG Yongji;CAO Huimin;ZHANG Li;ZHENG Dongyun;XIE Qinlan(College of Biomedical Engineering&Key Laboratory of Cognitive Science of State Ethnic Affairs Commission&Hubei Key Laboratory of Medical Information Analysis and Tumor Diagnosis&Treatment,South-Central Minzu University,Wuhan 430074,China)

机构地区：[1]中南民族大学生物医学工程学院&认知科学国家民委重点实验室&医学信息分析及肿瘤诊疗湖北省重点实验室,武汉430074

出　　处：《中南民族大学学报（自然科学版）》2022年第4期454-458,共5页Journal of South-Central University for Nationalities：Natural Science Edition

基　　金：国家自然科学基金资助项目(61501526,61178087);中央高校基本科研业务费专项资金资助项目(CZQ18014)。

摘　　要：采用近红外光谱法对血红蛋白进行定量分析时,数据集划分与预处理方法直接影响定量模型的预测精度.以120份不同浓度血红蛋白仿体溶液的光谱数据为研究对象,研究随机法、间隔划分法、KS法、Duplex法、SPXY法5种数据集划分方法对PCR、PLS定量分析模型预测精度的影响;研究27种预处理方法对PCR、PLS定量分析模型预测精度的影响,预处理组合时考虑组合顺序的影响因素.实验结果表明:PLS模型最优的数据集划分方法是SPXY法;PCR模型最优的数据集划分方法是间隔划分法.27种预处理方法,PLS模型和PCR模型优选出的最佳预处理方式均为SG+DOSC.此时,PLS模型的RMSEP值为3.5532;PCR模型的RMSEP值为14.9032.研究结果为此类光谱数据的处理提供了一种思路和方法.Near infrared spectroscopy(NIRS)was used to quantitatively analyze hemoglobin concentration.The methods of dataset partitioning and pretreatment directly affect the prediction accuracy of the quantitative model.Taking the spectral data of 120 hemoglobin biomimetic solution with different concentrations as the research object,the effects of five dataset partitioning including random method,interval partition method,KS method,duplex method and SPXY method on the prediction accuracy of PCR and PLS quantitative analysis models were studied.The effects of 27 pretreatment and combined methods on the prediction accuracy of PCR and PLS quantitative analysis models were studied.Experimental results indicate that:The optimal dataset partitioning of PLS model is SPXY method,and the optimal dataset partitioning of PCR model is interval partition method.Among the 27 pretreatment methods,the optimal pretreatment method for PLS and PCR model is SG+DOSC.Then the RMSEP value of PLS model and PCR model was 3.5532 and 14.9032 respectively.The results provide an idea or method for the processing of this type of spectral data.

关 键 词：数据集划分 预处理方法 近红外光谱 定量模型 

分 类 号：O433.4[机械工程—光学工程]