Estimation of blood glucose level based on PPG signals measured by smart devices

Abstract

This paper deals with the possibilities of non-invasivedetermination of blood glucose from photoplethysmographic signals.Monitoring blood sugar is the most important part of managingdiabetes. Diabetes is one of the world’s major chronic diseases.Untreated diabetes is often a cause of death.Two datasets have been created by recording thephotoplethysmographic signals of 16 people using two smart devices(a smart wristband and a smartphone), along with their bloodglucose levels measured in an invasive way. Thephotoplethysmographic signals were preprocessed, and suitablefeatures were extracted from them. The aim of the work is to proposemethods for glycemic classification and prediction.Various machine-learning models were created. The best modelfor classifying blood glucose into two groups (low blood glucose andhigh blood glucose) is random forest, which achieves an F1 score of84% and 80% on two different test sets obtained from two smartdevices. The best blood glucose level prediction model is also basedon random forest and achieves an MAE of 1.02 mmol/l and 1.17mmol/l on both testing datasets.