基于高光谱技术的细菌生物被膜分类检测  

Classification and detection of bacterial biofilms based on hyperspectral fluorescence imaging

作　　者：牛晓虎 冯耀泽[1,2,3,4] 鲍雪 崔恒洁 王梦冉 岑晓旭 孙光全 NIU Xiaohu;FENG Yaoze;BAO Xue;CUI Hengjie;WANG Mengran;CEN Xiaoxu;SUN Guangquan(College of Engineering,Huazhong Agricultural University,Wuhan 430070,China;Key Laboratory of Aquaculture Facilities Engineering,Ministry of Agriculture and Rural Affairs,Wuhan 430070,China;Agricultural Equipment Laboratory of the Middle and Lower Reaches of the Yangtze River,Ministry of Agriculture and Rural Affairs,Wuhan 430070,China;Shenzhen Institute of Agricultural Genomics,Chinese Academy of Agricultural Sciences,Shenzhen 518000,China;College of Animal Science and Technology,College of Veterinary Medicine,Huazhong Agricultural University,Wuhan 430070,China)

机构地区：[1]华中农业大学工学院,武汉430070 [2]农业农村部水产养殖设施工程重点实验室,武汉430070 [3]农业农村部长江中下游农业装备实验室,武汉430070 [4]中国农业科学院深圳农业基因组研究所,深圳518000 [5]华中农业大学动物科学技术学院、动物医学院,武汉430070

出　　处：《华中农业大学学报》2023年第3期241-249,共9页Journal of Huazhong Agricultural University

基　　金：中央高校基本科研业务费专项(2662020GXPY003);华中农业大学深圳营养与健康研究院研发项目(SZYJY2021028)。

摘　　要：针对现有生物被膜检测方法耗时、费力、低效的问题,以大肠杆菌、金黄色葡萄球菌、沙门氏菌为例,研究荧光高光谱技术对不同细菌生物被膜进行种类识别和成膜能力评价的可行性。采集细菌生物被膜样本荧光高光谱图像,并基于5种方法预处理后的光谱数据建立支持向量机分类(support vector classification machine,SVC)和偏最小二乘判别分析(partial least squares discriminant analysis model,PLS-DA)细菌被膜分类检测模型。利用连续投影算法(successive projections algorithm,SPA)、竞争性自适应重加权算法(competitive adaptive reweighted sampling,CARS)分别提取特征波长并建立相应简化模型。结果显示:细菌生物被膜种类识别全波长和特征波长模型中SVC性能均优于PLS-DA,最优模型为None-SPA-SVC,校正集和预测集分类准确率均为96.67%。在细菌生物被膜成膜能力的全波长模型分类判别中,SVC算法整体上分类准确率优于PLS-DA;对于简化模型,最优模型为SPA-SVC,校正集和预测集分类准确率分别为100.00%和96.67%。研究结果表明,高光谱技术可以对细菌生物被膜种类及生物被膜的成膜能力进行有效、快速、准确地分类。Bacterial biofilms widely exist on the surfaces of food processing machinery,medical equipment and the environment,and bring a huge threat to human health due to strong drug resistance.Escherichia coli,Staphylococcus aureus,and Salmonella typhimurium were used to study the feasibility of species identification and evaluate film-forming ability of different bacterial biofilms by hyperspectral fluorescence imaging technology to solve the problems of time-consuming,laborious and inefficient detection of existing biofilms.The hyperspectral fluorescence images of bacterial biofilm samples were collected,and support vector classification machine(SVC)and partial least squares discriminant analysis(PLS-DA)models based on the spectral data preprocessed by five methods were established to classify bacterial biofilms.Characteristic wavelengths were extracted by employing successive projections algorithm(SPA)and competitive adaptive reweighted sampling(CARS)respectively,and the corresponding simplified models were established.The results showed that SVC outperformed the full-wavelength and characteristic-wavelength models of identifying bacterial biofilm species than the PLS-DA,with the optimal model being None-SPA-SVC,where the classification accuracy of calibration set(CCRC)and prediction set(CCRP)were both 96.67%.In the classification and discrimination of film-forming ability of bacterial biofilm,the full-wavelength SVC models generally outperformed PLS-DA with higher classification accuracy.For the simplified models,the optimal model was SPA-SVC,with CCRC and CCRP of 100.00%and 96.67%,respectively.It is indicated that hyperspectral fluorescence imaging technology can effectively,quickly and accurately classify the types of bacterial biofilms and the film-forming ability of biofilms.

关 键 词：高光谱技术 生物被膜 成膜能力 偏最小二乘分类 支持向量机分类 

分 类 号：Q93-31[生物学—微生物学]