Fast Serum Detection Technology For Liver Diseases Based On Surface-enhanced Raman Spectroscopy And Multivariate Statistical Analysis

ObjectiveEarly detection and diagnosis of liver disease is of great significance for the treatment of liver disease,However,the existing detection methods have the disadvantages of long detection period,high cost,etc,This study is aimed at some deficiencies of the existing methods for early diagnosis of liver diseases,we explored differences in the surface-enhanced Raman spectroscopy(SERS)of blood serum to discriminate several types of chronic hepatitis B(CHB)patients(CHB,CHB-liver cirrhosis compensatory period,CHB-liver cirrhosis decompensatory period)and hepatocellular carcinoma(HCC)patients(Stages 0,1,2,and 3)from normal individuals and esophageal cancer(EC).The differences among various types of serum based on multivariate statistical analysis were analyzed.Diagnosis models for liver diseases was built,and a rapid and non-invasive detection technique for liver diseases based on serum SERS was established.Early diagnosis and prevention of liver diseases could be achieved.MethodsAn improved SERS substrate was selected by comparing the enhancement effect of serum Raman signals using Au and Ag nanoparticles.SERS signals of serum were obtained from 304 healthy individuals,48 patients with CHB,49 patients with CHB-liver cirrhosis(CHB-LC)compensatory period,48 patients with CHB-LC decompensatory period,46 patients with Stage 0 HCC,46 patients with Stage 1 HCC,49 patients with Stage 2 HCC,47 patients with Stage 3 HCC and 99 patients with EC.The spectroscopy result was analyzed after spectral smoothing,baseline subtraction,and normalization.The final Raman peak attribution was determined.The differences in serum SERS spectra among normal individuals,EC patients and patients with liver diseases,which included three types of CHB patients(CHB,CHB-LC compensatory period,and CHB-LC decompensatory period)and four types of HCC patients(Stages 0,1,2,and 3),were compared based on a supervised learning method,i.e.,orthogonal partial least squares discriminant analysis(OPLS-DA).The classification efficiency of each model was evaluated by analyzing the principal component score chart and the receiver operating characteristic(ROC)curve.10-fold cross-validation was conducted to evaluate the generalization capability of the models.The observations and the robustness of the models were assessed using 200 permutation tests.The variable important projection(VIP)of the models was analyzed for a good classification case.The Raman peaks that corresponded to Raman shifts of VIP > 1 were selected.The cause of the differences in metabolism among various types of liver diseases were determined.The differences in serum SERS spectra among normal individuals,EC patients and patients with liver diseases,including three types of CHB patients(CHB,CHB-LC compensatory period,CHB-LC decompensatory period)and four types of HCC patients(Stages 0–3),were compared based on unsupervised learning method,i.e.,principal component analysis(PCA)-CLASS.An independent PCA model for each packet was established according to different classification methods.The maximum retention degree of the original information of the sample was ensured.Coomans' plot was a direct reflection of the forecast grouping.The classification of the samples was assessed based on the orthogonal distance from the samples to the model.The overall performance of the classifier was evaluated based on the ROC curve.The 10-fold cross-validation method evaluated the generalization capability of the model to unknown samples with regard to the good classification model.ResultsThe enhancement effect of Ag nanoparticles on serum Raman spectra was 4–5 times stronger than that of Au nanoparticles.Therefore,Ag nanoparticles were used as SERS substrate in the later stage of this study.The differences in serum SERS spectra among various groups were compared.Differences were observed in the following Raman shifts: 556,638,724,760,811,853,888,958,1021,1095,1132,1218,1326,1339,1438,1580,and 1655 cm-1.These shifts belonged to tryptophan,lactose,acetyl coenzyme A,tryptophan,tyrosine,L-serine,glutathione,valine,phenylalanine,d-mannose,amide III,lipid,adenine,acetoacetate,and amide I.The SERS spectra of the serum obtained from normal individuals,EC patients and patients with liver diseases were compared based on OPLS-DA.The principal component score chart showed that the three types of samples exhibited a significant separation trend.The value of the integration area under the ROC curves(AUC)of normal individuals,EC patients and patients with liver diseases were 0.998,0.997 0.997,The correct rates of the training set and the test set of 10-fold cross-validation were 97.03% and 95.33%,respectively.The results showed that the model was robust in 200 permutation tests.The 12 Raman peaks obtained via VIP screening were significantly different between the three groups.These differences were caused by amino acids,carbohydrates,lipids,and other substances.The AUCs of normal individuals,EC patients and patients with liver diseases were 1,1,0.651 based on the comparison with PCA-CLASS,10-fold cross-validation results are as follows.The average correct rate of the training set was 74.02%,and that of the test set was 57.53%.The classification effectiveness of PCA-CLASS in patients with liver diseases,EC patients and normal individuals was lower than that of the supervised OPLS-DA.The SERS spectra of serum were compared between CHB patients and HCC patients based on OPLS-DA.The separation trend between the two groups was evident in the score chart.The value of the integration AUC of the two groups was 0.997.The average values of the training set and the test set of the 10-fold cross-validation were 96.33% and 94.27%,respectively.The results of the 200 permutation tests showed that the model was robust.Significant differences were observed between the two groups in Raman signal intensity at eight Raman shifts after VIP screening.These differences were associated with amino acid metabolism,glucose metabolism,and other metabolic activities.The AUC(CHB)was 0.765 and the AUC(HCC)was 0.740 based on the ROC of PCA-CLASS analysis.The average correct rates of the training set and the test set of 10-fold cross-validation were 96.00% and 85.00%,respectively.The two methods were feasible for clinical sample analysis compared with OPLS-DA.The differences among the SERS spectra of the serum obtained from CHB,CHB-LC compensatory period,and CHB-LC decompensatory period patients were compared based on OPLS-DA.The scores of the three types of patients were extended in three directions in the score chart,and they presented a significant classification trend.The AUCs of the CHB,CHB-LC compensatory period,and CHB-LC decompensatory period patients were 0.993,0.997,and 0.972,respectively.The average correct rates of the training set and the test set of 10-fold cross-validation were 91.00% and 87.00%,respectively.The permutation test results showed that the model was stable.A total of 14 significant differences were observed among the three groups in the Raman shifts via VIP screening.The results showed differences in the contents of glutathione,lipid,and tyrosine in different stages of liver cirrhosis.The AUCs of the serum SERS spectra of the CHB,CHB-LC compensatory period,and CHB-LC decompensatory period patients were 0.837,0.933,and 0.954,respectively,based on PCA-CLASS.The training set and the test set of 10-fold cross-validation had correct rates of 87.81% and 67.00%,respectively.The SERS spectra of the serum obtained from patients with four stages of HCC were compared based on OPLS-DA.The results showed that the four groups of patients with different HCC stages tended to separate in four directions.However,overlapping occurred in the middle position.The AUCs of the patients with Stages 0–3 HCC were 0.996,0.998,0.989,and 0.992,respectively.The average correct rates of the training set and the test set according to 10-fold cross-validation were 88.77% and 81.76%,respectively.The regression line intercept was normal after 200 permutation tests,thereby indicating that the model was robust.Subsequently,13 effective displacement values were selected according to the VIP value.These differences were caused by amino acids,carbohydrates,lipids,and other substances,and these different substances play key roles in a series of biochemical reactions,such as biological transformation,immune process monitoring,signal transduction,and nutrient metabolism.The ROC curves were drawn based on the unsupervised PCA-CLASS,and the following AUCs were obtained: Stage 0=0.857,Stage 1=0.756,Stage 2=0.809,and Stage 3=0.940.The correct rates of the training set and the test set according to 10-fold cross-validation were 78.71% and 54.85%,respectively,which showed that the generalization capability of the model was normal compared with the OPLS-DA model.ConclusionThis method based on SERS and two multivariate statistical analysis methods(OPLS-DA and PCA-CLASS)not only has a high accuracy,but also is fast and sensitive to the detection and classification of several kinds of liver diseases.The supervised OPLS-DA classification model exhibits good classification capability based on the comprehensive comparison of OPLS-DA and PCA-CLASS.The generalization capability of the model to unknown samples is robust.The two methods(OPLS-DA and PCA-CLASS)can be used to analyze unknown samples simultaneously.The preliminary study demonstrates that the approach based on multivariate statistical analysis can be used to supplement early diagnosis methods for liver diseases,such as the fast and non-invasive serum detection technology based on SERS.