Improving The Accuracy Of Raman Spectrum Medical Diagnosis Model By Advancing The Dimensionality Reduction Method

Raman spectroscopy can be used to characterize and explain changes of human tissue components from molecular level,which has certain clinical reference value for the diagnosis of diseases.Compared with conventional medical diagnostic methods,Raman spectroscopy is fast,accurate,simple and non-destructive.But Raman spectrum data obtained by Raman technology are large in size and noisy.It is difficult to directly identify diseases.To solve this problem,we must finish data preprocessing to establish a medical diagnosis model.The established diagnosis model can assist to clinical treatment.The main content of this study is to build Raman database,improve the dimensionality reduction method and improve the accuracy of classification models.In this thesis,510 breast samples were provided by Department of Breast Surgery in the First Hospital of Jilin University and 443 amniotic fluid samples were collected by Department of Gynecology in the First Hospital of Jilin University.Because of data noise,smoothing algorithm is applied to accomplish data preprocessing.The preprocessed Raman spectrum is high dimensional and small samples.In order to avoid over-fitting,an improved dimensionality reduction method(LPR)is proposed based on common feature selection and feature extraction methods.The LPR method is used to reduce the dimension of data.Support vector machine(SVM),limit learning machine(ELM)and K nearest neighbor(KNN)classification algorithm are built to predict the type of disease.A large number of experiments were carried out on breast cancer data sets and amniotic fluid data sets in order to prove the effectiveness of the LPR algorithm.LPR was compared with other single feature extraction or selection algorithms.The experimental results show that the combined improved dimension reduction method LPR with the SVM classification algorithm(LPRS)obtained the better classification accuracy than other models.The accuracy of LPRS recognition model for common diseases of pregnant women was 95.49%,the value of Kappa was 0.94,the accuracy of LPRS model for breast cancer was 83.33%,and the value of Kappa was 0.81.This indicated that the classification results of LPRS model had reference value.The results showed that the LPRS model established in this study had good recognition ability and could be used for the identification of common diseases of pregnant women and breast cancer.