Clinical evaluation of non-invasive blood glucose measurement by using near infrared spectroscopy via inter- and intra-subject analysis

Abstract

Introduction Non-invasive blood glucose measurement through near infrared (NIR) is possible but its reliability is questioned. The aim of the thesis is to evaluate the reliability of non-invasive blood glucose measurement by using NIR spectroscopy. Methods The experiments on glucose solutions provided positive results, which were followed by three clinical trials on healthy people and diabetic patients. The NIR transmittance method was applied on the first two clinical trials, while the NIR reflectance method was used on the last clinical trial, which was specifically designed for intra- and inter-subject comparisons. Partial least square (PLS) regression and preprocesses of piecewise, direct standardization (PDS) and Savitsky-Golay smoothing and differentiation (Savgol) were used for the glucose prediction and comparison. Results and Discussions A higher predictability of NIR glucose measurement was found from the spectra of the intra-subject when compared with the inter-subject. The R correlation coefficient of prediction (Rp) of inter-subject ranged from 0.73 to 0.80, while Rp of intra-subject were 0.90 to 0.94. The root mean square error of prediction (rmsep) ranged from 2.49 to 3.52mmol/l (inter-subject) and 1.49 to 2.10mmol/l (intra-subject). The Rp of intra-subject were 15 to 25% higher than that of inter-subject. The rmsep of intra-subject was 43 to 74% lower than that of inter-subject. In addition, the results also supported the proportion that a larger calibration sample size to prediction sample size ratio provided a better prediction, in which Rp and rmsep could reach 0.92 and 1.71mmol/l, respectively, for 4:1 ratio (calibration sample size-to-prediction sample size). The Savgol preprocess was clinically proven to be suitable for the NIR spectroscopic reflectance measurement of human fingers. The NIR method provides an instantaneous measure to the glucose in blood. However, the method may be affected by physiological influences over long-term measurements. Conclusions Non-invasive NIR blood glucose measurement was achievable for an instantaneous measure. The major effect for NIR spectroscopic measurement was the physiological difference. Therefore, the development should focus on intra-subject data acquisition which showed relatively low physiological influence. Sufficient sampling data for calibrations and prediction were also vital, with boosting by a suitable preprocess.