Intelligent Real-Time Hypoglycemia Prediction for Type 1 Diabetes

Hypoglycemia in Type 1 Diabetes (T1D) refers to a condition where blood glucose (BG) levels drop to abnormally low levels, typically below 70 mg/dL. This can occur when there is an excessive amount of insulin relative to the blood glucose level, leading to an imbalance that can be dangerous and potentially life-threatening if not promptly treated. The availability of large amounts of data from continuous glucose monitoring (CGM), insulin doses, carbohydrate intake, and additional vital signs, together with deep learning (DL) techniques, has revolutionized algorithmic approaches for BG prediction in T1D, achieving superior performance. In our study, we employed a Long Short-Term Memory (LSTM) neural network architecture to predict hypoglycemia events in patients with T1D. For the training and testing, we utilized the OhioT1DM (2018) dataset. In addition, real-time data collected from an individual patient for the evaluation. This patient utilized the CGM FreeStyle Libre (FSL) system, along with a smartwatch to monitor step count. The LSTM model exhibited performance demonstrating exceptional levels of sensitivity, specificity, and accuracy scores of 97.09%, 94.17%, and 95.63%, respectively, when assessed using the Ohio test dataset. Our research provides strong evidence supporting the system's efficacy in managing hypoglycemia events in individuals diagnosed with T1D.