A Non-invasive Blood Glucose Detection Method: Using Fluctuation Analysis to Evaluate Heart Rate Variability

Blood glucose (BG) detection is of great importance for diabetes prevention in healthy individuals. However, the traditional invasive detection methods have disadvantages, such as causing pain and inconvenience. Additionally, BG changes can disrupt the balance of the autonomic nervous system, which can be evaluated using heart rate variability (HRV). For these reasons, this study aims to develop a non-invasive BG level detection method based on HRV. This study proposed dynamic and static fluctuation analysis based on detrended fluctuation analysis (DFA) and multifractal detrended fluctuation analysis (MF-DFA). A physiological measuring system was employed to record the electrocardiogram and obtain R-wave interval (RR) time series during 40 oral glucose tolerance tests in 20 healthy individuals, and the corresponding BG concentrations were measured every 10 minutes. Besides, dynamic and static fluctuation analysis were used to extract HRV features from the RR series and compared with the classical frequency-domain method. Additionally, four machine learning models were applied to classify high and low BG levels to evaluate the effect of non-invasive BG detection. The results demonstrated that both dynamic and static features of DFA and MF-DFA could significantly distinguish different BG levels (p < 0.05). The accuracy, specificity, and sensitivity of the light gradient boosting machine were 80.0%, 74.5%, and 87.2%, respectively. Our model also displayed generalization potential on the D1NAMO dataset. The dynamic and static fluctuation analysis could evaluate the change in HRV under different BG levels. This novel method is a promising technique for non-invasive BG detection in healthy individuals.