| Raman spectroscopy technology has great application prospects in cancer diagnosis,biological research and materials by virtue of its unique sample molecular fingerprint information,as well as its advantages of non-invasiveness,high chemical specificity and low water interference.However,due to the complexity of Raman spectral information,it is challenging in the interpretation of sample information,so effective data analysis methods are needed to interpret complex spectral data.With the development of artificial intelligence technology,more and more machine learning algorithms are applied in Raman spectral data processing.For example,supervised learning algorithm for classification modeling and cluster analysis,spectral decomposition algorithm and Multivariate Curve Resolution-Alternating Least Squares can be used for Raman imaging.Multivariate analysis methods show great potential in sample information and structure analysis,and these methods also provide a theoretical basis for Raman spectral data analysis.In summary,the specific work of this paper is as follows:Firstly,the complexity of Raman spectroscopic data and the necessity of using data analysis methods are discussed,and the principles of spectral pretreatment methods,Raman spectroscopy and Raman spectroscopic imaging methods used in this paper are explained,and the model performance is evaluated by cross-validation methods and performance evaluation indicators,and the research scheme and data analysis methods of subsequent work are determined based on the above principles.Secondly,based on the Matlab platform,the design and implementation of Raman spectrum data analysis software(RSI-LAB),determine the workflow of the software platform,as well as the realization of cosmic ray removal,fluorescence background removal,Savitzky-Golay smoothing,normalization,mean centralization,one and two derivatives and other data pretreatment functions.The preprocessed data can be analyzed by principal component analysis,linear discriminant analysis,quadratic discriminant analysis,partial least squares discriminant analysis and support vector machine.Hierarchical clustering analysis,K-means clustering analysis,vertex component analysis,N-FIDNR algorithm,simplex growth algorithm,pure pixel index and MCR-ALS analysis are used for imaging.The software platform has a complete data analysis system,which can efficiently and conveniently process the biochemical information contained in the interpretation data of Raman spectral data.Finally,in order to test the performance of software platform,data preprocessing,Raman spectrum analysis and Raman spectrum imaging are carried out with actual samples.Firstly,the performance of two kinds of software(WITec and RSI-LAB)in data preprocessing is compared.Secondly,PCA-LDA,PCA-QDA,PLS-DA and PCA-SVM classification models were established by using the pretreated cell point spectral data of different doses of drug culture to effectively distinguish the spectral differences between the untreated,20μM and 40μM groups.Finally,the complete osteosarcoma cell data set was used to reconstruct the cell image based on clustering analysis algorithm(HCA and KCA),spectral decomposition algorithm(VCA,N-FINDR,SGA and PPI),and MCR-ALS imaging analysis.The spatial information of subcellular structure and biochemical components was interpreted by atlas combination.Based on the principle of machine learning algorithm,the software RSI-LAB is designed and implemented using Matlab GUI platform,and the potential of the software in biological Raman spectral data analysis was verified by biological samples.It provides effective data processing software for biological Raman spectroscopy research.However,the software platform has some shortcomings in interface interaction,and there is a gap between the pretreatment and commercial software in the removal of fluorescent background noise. |
PDF Full Text Request