基于支持向量机法的近红外光谱鉴别耐甲氧西林金葡菌和甲氧西林敏感金葡菌的研究  

Discrimination of methicillin-resistant and methicillin-susceptible Staphylococcus aureus using near-infrared spectroscopy based on Support Vector Machine method

作　　者：马宁[1] 李勇明[2] 尹美芳 张大勇[3] 王品[2] 骆永全[3] 龚雅利[1] 陈志强[1] 薛冬冬[1] 吴军[1] 

机构地区：[1]第三军医大学西南医院全军烧伤研究所、创伤、烧伤与复合伤国家重点实验室,重庆400038 [2]重庆大学通信工程学院通信工程系,重庆400044 [3]中国工程物理研究院流体物理研究所,四川绵阳621000

出　　处：《解放军医学杂志》2016年第12期1005-1009,共5页Medical Journal of Chinese People's Liberation Army

摘　　要：目的探讨使用近红外光谱结合支持向量机法区分耐甲氧西林金葡菌（MRSA）和甲氧西林敏感金葡菌（MSSA）的可行性。方法制作MRSA和MSSA的浓度标准曲线。扩增待测细菌，并根据公式制备相同浓度菌液。采集菌液样本近红外光谱数据，并对数据进行一阶求导、平滑去噪、归一化和基线校正等预处理。根据两种细菌光谱曲线的相关性，对900-2200nm波段数据进行主成分分析。依据累计贡献率结果，选择前三个主成分作为支持向量机的输入向量，分别使用线性、多项式、径向基3种核函数进行建模，比较不同模型区分MSSA和MRSA的准确性。结果MRSA和MSSA预处理后的光谱曲线相关系数为1．000，两者高度相似。使用主成分处理并采用3种支持向量机核函数建模后，模型的训练和测试准确率均高于95％，其中采用径向基核函数分类结果最好，训练准确率为99．72％±0．21％，测试准确率为99．47％±0．00％。结论使用近红外光谱结合支持向量机的分析方法具有精确区分MRSA和MSSA的能力。Objective To explore the feasibility of distinguishing between methicillin-resistant S. aureus （MRSA） and methicillin-susceptible Staphylococcus aureus （MSSA） using near-infrared spectroscopy and the Support Vector Machine analysis method. Methods The concentration standard curve of the MRSA and MSSA was prepared. The bacteria were amplified and prepared into the same concentration according to the concentration formula. Near-infrared spectroscopy data of the MRSA and MSSA were collected and pretreated with first derivative, smoothing, normalization and baseline correction. After the pretreatment, the correlation analysis of the two kinds of bacteria was executed. The spectral data of bacteria with the wavelength from 900 to 2200 nm were analyzed by principal components. According to the results of cumulative contribution rate, the first three principal components were extracted and used as the input vector to establish Support Vector Machine models in three classifiers （linear, polynomial and RBF） and then a comparison of the three models was performed. Results The correlation coefficient of the pretreatment spectral curve of MRSA and MSSA was as high as 1.000. The training and test accuracies of the models were all over 95% after using the principal component analysis and Support Vector Machine models in three classifiers （linear, polynomial and RBF）. The RBF classifier had the highest accuracy, and the result was 99.72%± 0.21% for training accuracy and 99.47% ±0.00% for the test accuracy. Conclusion Near-infrared spectroscopy and the Support Vector Machine analysis method has a high ability to discriminate between MSSA and MRSA.

关 键 词：谱学 近红外线 耐甲氧西林金葡菌 主成分分析 支持向量机 

分 类 号：R372[医药卫生—病原生物学]