Makine öğrenimi ile EKG, PPG ve uzak PPG sinyalleri kullanarak glukoz seviyesi tespiti


Çakın Ö., Karaman O.(Executive), Gürpınar E., Polat Ö., Sarı R., Aydemir M., et al.

TUBITAK Project, 2022 - 2024

  • Project Type: TUBITAK Project
  • Begin Date: November 2022
  • End Date: April 2024

Project Abstract

Diabetes is considered as one of the most important health problems due to its increasing number and frequent complications. Diabetes is a lifelong disease caused by the deficiency or ineffectiveness of the hormone insulin. The aim of diabetes treatment is to keep the quality of life of the person high by providing blood sugar regulation and to prevent the development of long-term complications. Diabetes, a metabolic syndrome, requires frequent follow-up and leads to serious complications and even death if not managed properly. With the developing technology, new health approaches are used in the follow-up of individuals with diabetes and in preventing complications. Electrocardiography (ECG) is obtained by recording the electrical activity occurring in the heart to examine the functioning of the heart muscle and the nervous conduction system. Photoplethysmography (PPG) is an optical technique used to detect volumetric changes in blood in the peripheral circulation. PPG simply works on the principle of illuminating the skin and measuring changes in light absorption. In recent years, machine learning models have begun to be developed in which glucose levels are detected non-invasively using PPG and ECG signals. In order to increase the accuracy of these studies, which have shown great success with machine learning, it is planned to increase the accuracy by examining both ECG and PPG signals at the same time and to carry out a pioneering study in the field. In the literature researches, it has been seen that the glucose level in the blood can be determined by ECG and PPG signals. In these studies, it has been suggested that there is a functional relationship between PPG signal and glucose levels. It has been suggested that this functional relationship is related to the hemodynamics of the individual, in addition, the autonomic nervous system partially reflects the state to PPG and EKG signals, and that the glucose level changes the blood viscosity, which will affect the pressure change in the vessels, and this relationship can be confirmed by machine learning. In the light of this information, one of the hypotheses put forward within the scope of the project is that the force that the heart can exert on the vessels depending on the glucose change changes as a function of the frequency, based on the fact that the physical properties of the circulatory system change with the glucose level in the blood, and that the responses in the frequency space will change when approached with an analogy in the form of a mass-spring system. Therefore, it is predicted that the glucose change will cause the electrical potential to change during the contractions caused by the depolarization of the heart, and therefore the ECG signal will undergo a frequency-dependent change. Remote PPG (rPPG) is the remote detection of color changes in the skin by the blood filling of the capillaries and arteries in the dermis and epidermis layers of the skin. rPPG signals are obtained by monitoring the reflection of an external light source on the human body with a camera. Considering the successes achieved by analyzing only PPG signals in literature research, it was hypothesized that a similar performance could be achieved with rPPG. Since the working principle of rPPG is very similar to the working principle of reflective pulse oximetry devices, it can be said that an estimation of the amount of glucose in the blood can be made with machine learning only by analyzing the rPPG signals. In this context, within the scope of the project, the estimation of glucose level will be provided by the analysis of rPPG signals only through the camera, and it will be a pioneering study in this field. Akdeniz University Department of Endocrinology and Metabolic Diseases and Notrino Research Bil-Tek Ar-Ge LTD. ŞTİ. It is planned to develop artificial intelligence models that predict glucose by taking ECG, PPG and video recordings from a total of 3000 participants during the project period, which will be managed by the project. Since the proposed project is within the scope of priority areas in the national and international arena, it is anticipated that the project outputs will contribute significantly to the literature in the national and global sense. It is planned to publish studies in respected journals scanned in the International Science Citation Index. As a result of the successful completion of the project, it is aimed to develop high-performance, high-quality, cost-effective and reliable solutions with the prototype to be produced, and to brand it by offering reliable products and solutions with high added value.