Relationship between PPG Signals and Glucose levels through Chaotic Descriptors and Support Vector Machines

Abstract

When shining a light through a finger, some of it will be absorbed by oxygenated and unoxygenated hemoglobin. Measuring the absorbed light over time provides the photo-plethysmographic (PPG) waveform, which can represent the blood flow of a subject. One way of obtaining the PPG waveform is to use the camera and flash of a smartphone, placing them on the finger of a subject, and analyzing the variation of red color. The PPG can also be obtained using oximeter-like devices, which are non-invasive and safe. In contrast, to measure the blood glucose of a subject, a glucometer is used, which is a device that is typically invasive and expensive. Therefore, we propose the use of the following descriptors from Chaos theory to analyze the PPG signal: correlation dimension, maximum Lyapunov exponent and Hurst exponent. Then, these values are converted into a 3-dimensional vector that can be represented in a 3-dimensional space. Each vector has an associated glucose level that is used to train an algorithm which classifies all the vectors in three different ranges of blood glucose levels.