Utilization of IMU-Based Gesture Recognition in the Treatment of Diabetes

Changes in blood sugar levels are significantly affected by dietary carbohydrate intake, which clearly affects insulin therapy. However, currently available insulin pumps do not perform continuous blood glucose monitoring, and the patient should do so. The quality of the treatment can be further improved if the phenomenon of eating can be recognized based on the characteristic movements, and then by sending an automatic signal of the event to the insulin pump, it can be taken into account that an increase in blood sugar levels is expected. In our work, we focused on this task and, using 4 different tools, collected data from more than 8 people on their hand movements during their meals, on which we performed various data manipulations. Finally, using the aggregation of the obtained data set, we tested different machine learning techniques, from which we will select the optimal algorithm that can recognize the motion characteristic of carbohydrate intake independently of the person with the smallest error. The article presents the tools used for the measurement, the operations performed on the data, as well as the tried and tested machine learning methods and their comparative analysis. Our goal is to improve the quality of life of people with diabetes by increasing the effectiveness of base bolus therapy.