Application Of Fourier Transform Infrared Spectroscopy Combined With Classification Recognition Algorithm In Screening High-risk Human Papillomavirus

Persistent infection with HR-HPV can easily lead to cervical cancer.Timely detection of HR-HPV and early treatment can effectively reduce the incidence of cervical cancer.FT-IR is a special mid-infrared spectrum.FT-IR has the advantages of fast diagnosis speed and high diagnosis accuracy,and can provide a variety of information about molecular changes,so it has been widely used in the field of biomedical science.Based on this,we collected the FT-IR of exfoliated cervical cells from 50 HR-HPV patients and 50 normal subjects,then diagnosed these spectra data by classification recognition algorithm to screen HR-HPV.In this thesis,we mainly study the following three aspects:1.HR-HPV screening was studied by FT-IR technique combined with PCA-LDA classification model.First,the collected original spectra were pretreated by MSC and z-score standardization,and then the PCA-LDA diagnostic model was established.The accuracy,specificity and sensitivity of the diagnostic results were 98%,98%and 98%,respectively,and AUC of the ROC curve was 0.997,which indicates that the PCA-LDA classification model could be used to screen HR-HPV.2.HR-HPV screening was studied by FT-IR technique combined with PCA-SVM classification model.We chose three kinds of kernel functions among svm algorithms,and established the corresponding PCA-SVM classification models,respectively.In the case of linear kernel function,the first 5 PCs wrer required to achieve the best diagnostic results.The diagnostic accuracy,specificity and sensitivity were 99%,98%and 100%,respectively,and AUC of the ROC curve was 0.99.In the case of polynomial kernel function,the best diagnostic results could be achieved when the first 7 PCs were taken.The diagnostic accuracy,specificity and sensitivity were 99%,98%and 100%,respectively,and AUC of the ROC curve was 1.In the case of gaussian radial kernel function,only the first 3 PCs could be selected to achieve the best diagnostic results,and the classification accuracy,specificity and sensitivity could be as high as 99%,98%and 100%.In this case,AUC of the ROC curve was 1.Therefore,the diagnostic effect of SVM is good when the three kinds of kernel functions are selected,but the diagnostic effect is better when the gaussian radial kernel function is selected.3.HR-HPV screening was studied by FT-IR technique combined with PCA-KNN classification model.When the first three PCs were selected,the accuracy,specificity and sensitivity of the PCA-KNN model diagnosis results could reach 99%,98%and 100%,respectively.AUC of the ROC curve was 0.99,which indicates that the model had good classification and diagnosis performance.In addition,the changes of biochemical substances corresponding to the peaks of the two groups of infrared spectra are consistent with previous related studies,which indicates that the spectra we collected are true and reliable.In conclusion,FT-IR combined with PCA-LDA,PCA-SVM and PCA-KNN can obtain good classification results,respectively.However,the PCA-SVM and the PCA-KNN classification model are slightly higher than the PCA-LDA in the accuracy,specificity and sensitivity.The above studies indicate that FT-IR combined with classification recognition algorithm in HR-HPV screening has a great prospect.