| Cancer seriously threatens the life of human.Breast cancer and lung cancer are the cancers with the highest incidence and the greatest number of deaths around the world,respectively.However,unified screening markers with high sensitivity and accuracy are still not available for different tumors.Therefore,it is urgent to develop a flexible screening method with high detection sensitivity.Surface Enhanced Raman scattering(SERS)technology is capable of sensitively obtaining the rich fingerprints of the sample or the Raman probe molecules during detection.By comparing the difference of the SERS spectrum of serums between normal and cancerous cell tissues,references can be provided for the diagnosis of cancer.Machine learning can transform the matter identification problem of Raman spectrum into a classification problem,which can meet the rapid detection of different target substances in complex systems.In this thesis,serum SERS spectrum was combined with principal component analysis(PCA),support vector machine(SVM)and convolutional neural network(CNN),which are three machine learning algorithms,to identify breast cancer and lung cancer.The main research contents of this thesis are as follows:1.Nano-silver glue was prepared by microwave method and was used as SERS substrate for the spectral collection of 132 serum samples(41 of healthy people,48 of breast cancer patients,and 43 of lung cancer patients).And the serum SERS spectrum was pretreated using the direct baseline method.2.Investigation on classification problems of SERS spectrum of cancer via the PCA-SVM method.Firstly,the spectral features were extracted by PCA.And then the normalized spectral features were classified by SVM based on linear kernel function and Gaussian radial basis(RBF)kernel function.The accuracy and average F1 value of the linear PCA-SVM algorithm were96.3% and 100%,respectively.The sensitivity of breast cancer and lung cancer were 90% and100%,respectively.The accuracy,mean F1 value and sensitivity of breast cancer and lung cancer of the RBF PCA-SVM algorithm were 100%.3.Investigation on classification problems of SERS spectrum of cancer via the CNN method.Data enhancement method was used to increase the SERS spectrum data of serum after straightening the baseline to 1000 groups,which were divided into the training set,the validation set,and the test set according to the ratio of 7:1:2.A 5-layer one-dimensional convolutional neural network was built under the Tensor Flow framework to extract and classify the spectral data of serum.The network comprised two convolutional layers,two fully connected layers,and one Soft Max classification layer.Additionally,it introduced the batch normalization,Adam optimizer,and dropout algorithm to optimize the network.Use the data groups to train and test the model.The accuracy of the training set,the validation set,and the test set were 99.71%,100% and 99.5%,respectively.The sensitivity and specificity of breast cancer and lung cancer were higher than 98%,and the F1 value of healthy people,breast cancer and lung cancer were more than 99%. |
