Three-wavelength PPG and support vector machine for non-invasive estimation of blood glucose

Abstract

Abstract Diabetes mellitus is a chronic disease characterized by producing abnormal levels of blood glucose concentration. Currently, the most widely accepted method for glucose monitoring is invasive, however, despite its great reliability, it can be uncomfortable and traumatizing for the youngest users. The objective of this study is to provide an alternative method that allows a non-invasive estimation of blood glucose levels with an elevated level of confidence. In this work, 187 records were performed on people without any declared pathology; the concentration of blood glucose and the amplitude of the PPG signals of 525 nm, 660 nm and 940 nm were measured simultaneously. 70% of the data was used to train a regression model based on a fine Gaussian support vector machine, while the remaining 30% is used to validate the results. The regression model using the support vector machine was able to locate 95.38% of the estimates with an error of less than 15%, showing a standard error of 7.01 mg/dL and a MARD of 6.99%. The model presented here allows non-invasive estimation of blood glucose levels with reliability comparable to minimally invasive devices currently on the market.