| In recent years,pattern recognition technology develops rapidly,and the technology of detecting samples by Raman spectroscopy is becoming more and more mature,making the combination of the two in the biomedical field more and more widely used.At home and abroad,there have been a lot of related researches in the screening of various diseases such as cancer,thyroid,hepatitis C,dengue fever and skin diseases through the Raman spectra of blood,sweat and tissues,but there are few reports on the classification and recognition of ophthalmic diseases through the Raman spectra of tear.This paper analyzes the diagnosis of ophthalmic diseases and the research status of combining with Raman spectrum signals at home and abroad,and finds that pattern recognition algorithms,including Support Vector Machine(SVM),and Back Propagation Neural Network(BPNN),have great scientific research value in medical diagnosis and recognition.According to the characteristics of tear Raman spectrum signals of ophthalmic diseases,SVM and BPNN algorithm are used to classify and recognize respectively by combining PLS and PCA feature extraction.In this paper,adaptive iteratively reweighted Penalized Least Squares(airPLS)is used to further background subtraction of Raman spectrum,and the classification effect is improved by combining SVM and BPNN models with airPLSPLS and airPLS-PCA algorithm respectively.The main research contents of this article are the following two points:1.The common eye diseases are screened and diagnosed by using the collected tear Raman spectrum signals and different pattern recognition algorithms.In this paper,polynomial fitting noise reduction processing is carried out to extract the pure spectrum signal and normalization processing is carried out to make all the Raman spectrum signals in the same dimension.The PLS algorithm and PCA algorithm are used to extract the characteristics of all the tear Raman spectra.Finally,PLS-SVM,PCA-SVM,PLS-BPNN and PCA-BPNN models were established and compared.The accuracy of the four models is 85.71%,60%,84.29(± 1.35)% and 68.29(± 4.58)% respectively.The specificity and sensitivity of the normalized PLS-SVM model were 100% and 76.19%,respectively.The results show that the pattern recognition algorithm based on tear Raman spectrum has great potential in ophthalmic disease screening.2.In the third chapter,the accuracy of the model based on polynomial fitting noise reduction and normalization preprocessing is not ideal.The SVM and BPNN classification models based on airPLS-PLS and airPLS-PCA are proposed,and Learning Vector Quantization(LVQ)neural network is used as the comparison algorithm.Finally,the average accuracy of the six classification models,airPLS-PLS-SVM、airPLS-PLSBPNN、airPLS-PLS-LVQ、airPLS-PCA-SVM、airPLS-PCA-BPNN and airPLS-PCALVQ is 100%,99.41%,79.41%,73.52%,79.41% and 73.52%,respectively.The results show that the improved pattern recognition algorithm has a better effect in ophthalmic disease screening.A large number of experimental results in this paper show that the combination of PLS-SVM、PLS-BPNN models can effectively identify eye diseases or healthy tears.And the SVM and BPNN algorithm optimized by airPLS-PLS and airPLS-PCA methods can improve the accuracy and performance of the classification model,which has great application value.It also provides a new idea for clinical diagnosis and screening of eye diseases. |
