Study On Multivariate Statistical Analysis And Assistant Diagnostic Methods Of Raman Spectroscopy In Leukemia Cells

Acute leukemia(AL)is one of the most common heterogeneous malignant hematologic disease in adults,with high incidence and high recurrence rate.The current diagnosis of AML is mainly made by morphological evaluation of bone marrow and peripheral blood.Further classification should be based on immunology,cytogenetics,and molecular biology Conventional AL diagnostic methods are highly subjective,expensive,time-consuming and cell-destructive.In addition,the relationship between biochemical indicators and the prognosis of AL is unclear,limiting its diagnosis and therapeutic decisions.Raman spectroscopy,as a detection technology based on photon inelastic scattering,has the characteristics of fast,label-free and non-destructive.It can reflect the vibration information of chemical bonds in the sample,and then analyze the type and structural information of substances contained in the sample at the molecular level,which has broad application prospects in the biomedical field.At present,for complex clinical leukemia samples,problems such as multi-cell mixing,interference of glass Raman signal and fluorescence effects will have a huge impact on subsequent leukemia Raman detection and classification.Therefore,it is very important to establish an adaptive cell collection and pretreatment method.Secondly,we analyzed the blood samples from different newly diagnosed patients.Aiming at the problem that the prognosis of AL is difficult to predict,a multi-angle Raman model was established,which can be used to from the diagnosis of leukemia,the classification of AML subtypes to the evaluation of chemotherapy effect and gene mutation identification.Although the model has high classification accuracy,its physical meanings are usually not well explained.Therefore,we further propose to combine multivariate curvature resolution-alternating least squares method to explore intracellular molecular changes and supplement the molecular interpretation of the Raman diagnostic model.The specific research contents are as follows:(1)Investigate the collection and pre-treatment methods applicable to clinical leukemia samples.For clinical blood smears,the influence of experimental processes such as sample preparation,spectroscopic instrument configuration and signal acquisition on the spectral quality was analyzed.And a standard clinical leukemia sample collection method was established to obtain high-intensity Raman spectral signals while achieving accurate screening of primary leukemia cells.Secondly,the interference signals present in the spectra were investigated.The Savitzky-Golay filter was used to remove the noise.The specific scale analysis method based on multiresolution wavelet transform was used to separate the substrate.The fluorescence background was processed by cubic spline curve interpolation fitting.A standard signal processing method was established to provide high-quality spectral data for subsequent diagnostic models.(2)A clinical acute leukemia classification model based on multivariate statistical methods was established to achieve leukemia diagnosis based on Raman spectroscopy.In this study,a large number of single-cell spectral data of AL patients were collected using a lab-built micro-Raman spectroscopy system.Principal component-linear discriminant analysis(PCA-LDA)algorithm was selected for the big data study of Raman spectroscopy,to develop and model the leukemia Raman spectroscopy data from multiple perspectives and levels from diagnosis of leukemia,subtype classification of AML,prognostic stratification assessment and gene mutation identification,with accuracy rates of 94.85%,91.62%,84.64% and 93.60%,respectively.By combining microfluidic microarray technology,an objective,nondestructive and highly accurate identification method of clinical leukemia cells was established,which has great potential for the rapid diagnosis and precise treatment of clinical acute leukemia.(3)Analysis of leukemia Raman spectra and classification models from the molecular level.To compensate for the deficiencies of PCA algorithm in terms of missing physical meaning and difficult interpretation of classification criteria,this paper uses multivariate curvature resolution-alternating least squares(MCR-ALS)to resolve the Raman fingerprint spectra of acute leukemia cells,decompose the spectral data into proteins,nucleic acids,and lipids(sugars)without a priori information,and explore the differences in their concentrations among different cell types to solve the problem of overlapping and confounding of spectral peaks.By correlating Raman features with the molecular mechanism of leukemia,the characterization of abnormal changes of macromolecules during cell carcinogenesis,drug resistance and gene mutation was achieved,and the Raman feature markers that facilitate classification were found.In summary,this paper has achieved a label-free and non-destructive study of leukemia cells at the molecular level through the analysis of Raman fingerprint data of cells from patients with initial diagnosis of AL,laying the foundation for a comprehensive exploration of the pathogenesis and therapeutic targets of AL in combination with multi-omics studies in the future.