Diagnosis Of Type Ⅱ Diabetes Based On ATR-FTIR Spectroscopy And XGBoost

Type Ⅱ diabetes has become a serious public health problem in the worldwide.it is impossible to achieve large-scale batch screening since the current clinical diagnostic methods have the shortcomings of complicated and time-consuming,which makes the patients with type Ⅱ diabetes fail to intervene in time and lead to the death,and a large number of pre-diabetes patients naturally turn into type Ⅱ diabetic patients.In this study,in order to achieve real-time,rapid and accurate diagnosis of type Ⅱ diabetes and its pre-patients,the ATR-FTIR spectroscopy for human serum and whole blood samples combined with XGBoost ensemble learning algorithm are used to establishing an optimal disease prediction model.The main research work of this paper is as follows:1.Diagnosis of type Ⅱ diabetes based on serum samples.Spectra of serum samples from 62 patients with clinical confirmed type Ⅱ diabetes and 55 healthy volunteers were acquired using ATR-FTIR.First,the CART algorithm is used to discriminate the ATR-FTIR data of serum samples.Then,the GS algorithm is used to optimize the parameter space to improve the robustness of the model.The accuracy of the established CART model is 80.85%.In order to improve the classification accuracy of the model,the XGBoost algorithm is used to discriminate,and the best CART model was used as the base classifier,the accuracy of the optimal XGBoost model was 93.61%.2.Diagnosis of type Ⅱ diabetes based on whole blood samples.Spectra of whole blood samples from 51 patients with clinical confirmed type Ⅱ diabetes and 62 healthy volunteers were acquired using ATR-FTIR.First of all,Considering the low purity of whole blood samples,the influence of different Savitzky-Golay smoothing modes on the effect of the model is discussed,and the original spectral data is replaced by the spectral data processed by the best Savitzky-Golay smoothing mode for modeling.Then,the CART algorithm is used to discriminate the processed data of the whole blood samples.The accuracy of the established SG-CART model is 82.60%.Finally,XGBoost algorithm is used for further discrimination,and the accuracy of the optimal SG-XGBoost model is 95.65%.3.Diagnosis of type Ⅱ pre-diabetes based on whole blood samples.Spectra of whole blood samples from 50 patients with clinical confirmed IFG,IGT and 62 healthy volunteers wereacquired using ATR-FTIR.First,the XGBoost combined with Savitzky-Golay smoothing algorithm is used to discriminate the ATR-FTIR data of the whole blood samples,and the accuracy of the optimal SG-XGBoost model is 91.5%.In order to improve the accuracy,the influence of PCA and ICA algorithm on the model’s performance is discussed.Finally,the spectral data processed by the best Savitzky-Golay smoothing mode and feature extraction algorithm are modeled.The accuracy of the established SG-ICA-XGBoost model is 97.77%.The results of this paper show that using ATR-FTIR spectroscopy combined with XGBoost algorithm for human serum and whole blood samples is feasible for real-time,rapid,and accurate diagnosis of type Ⅱ diabetes and pre-diabetes.