PROJECT TITLE :

Prediction of Adverse Glycemic Events from Continuous Glucose Monitoring Signal

ABSTRACT:

The most important goal of diabetes treatment is to keep blood glucose levels in the euglycemic range, preventing or at least reducing crucial hypo/hyperglycemic episodes. Modern continuous glucose monitoring (CGM) devices promise to give patients a better understanding of their glycemic situations when they approach dangerously low or high blood sugar levels. The objective is to recognize trends that lead to potentially dangerous situations in a timely manner, allowing the patient to make treatment options based on future (predicted) glucose concentration levels. We emphasize that a technical performance comparison of the alternatives offered in recent years has yet to be completed, therefore it is uncertain which is preferred. The goal of this research is to close this gap by comparing the most commonly used approaches for glucose event prediction. Both regression and classification methods, as well as static and dynamic training approaches, have been constructed and examined. The dataset includes 89 CGM time series from diabetic participants over the course of seven days. Performance metrics have been created and examined, which are especially defined to assess and compare the approaches' event-prediction capabilities. Our numerical results suggest that a static training technique performs better, especially when regression approaches are taken into account. When trained on a specific event category, such as hyperglycemia, classifiers achieve performance comparable to regressors, with the advantage of being able to predict events sooner.