Pre-diabetes diagnosis based on ATR-FTIR spectroscopy combined with CART and XGBoots

Abstract A rapid diagnosis model of pre-diabetes was established by using Attenuated total reflectance (ATR)-Fourier transform infrared spectroscopy (FTIR) combined with Classification and Regression Trees (CART) and eXtreme Gradient Boosting (XGBoost) ensemble learning algorithm. First, we collected peripheral blood samples of 112 volunteers and then measured the fasting blood glucose and 2h blood glucose levels of the oral glucose tolerance test, and determined the control group and the disease group according to the WHO diagnostic criteria (including impaired fasting glucose (IFG). Impaired glucose tolerance (IGT).In addition, ATR-FTIR spectra of peripheral blood samples were collected at the same time. The whole spectrum mid-infrared region (4000–600 cm −1 ) was used as the research object of diagnosis model of pre-diabetes. Second, preprocessing step, Savitzky-Golay (SG) smoothing pretreatment was performed and used PCA to extract spectral features. Finally, a rapid diagnosis model of pre-diabetes was established by using CART and XGBoost. CART model results were: Specificity: 80.00% (20/25), Sensitivity: 95.00% (19/20), Accuracy: 86.67% (39/45); XGBoost model results: Specificity: 100.00% (25/25), Sensitivity were: 85.00% (17/20)), Accuracy: 93.33% (42/45).The results show that the rapid diagnosis model of pre-diabetes was established by SG-PCA-XGBoost which has the best effect. Hence, the model established by ATR-FTIR spectra combined with SG-PCA-XGBoost in the experiment can more effectively diagnose pre-diabetes. This model needs no sample preprocessing; it is characterized by simple operation and time efficiency. Furthermore, the established model provides a fast accurate method for pre-diabetes diagnosis.