PROJECT TITLE :

Convolutional Recurrent Neural Networks for Glucose Prediction

ABSTRACT:

Blood glucose control is critical for diabetes management. Machine Learning techniques are used in current digital therapy approaches for people with type 1 diabetes, such as the artificial pancreas and insulin bolus calculators, to anticipate subcutaneous glucose for better control. Deep Learning has lately been used in Healthcare and medical research to obtain cutting-edge outcomes in a variety of tasks, including Disease Diagnosis and patient condition prediction. We present a deep learning model capable of forecasting glucose levels with leading accuracy for simulated patient cases (RMSE = 9.38 0.71 [mg/dL] over a 30-min horizon, RMSE = 18.87 2.25 [mg/dL] over a 60-min horizon) and real patient cases (RMSE = 21.07 2.35 [mg/dL] for 30 minutes, RMSE = 33.27 4.79 percent for 60 minutes). Furthermore, in both a simulated patient dataset (PH eff = 29.0 0.7 for 30 min and PHeff = 49.8 2.9 for 60 min) and a real patient dataset (PH eff = 19.3 3.1 for 30 min and PH eff = 29.3 9.4 for 60 min), the model provides competitive performance in providing effective prediction horizon (PHeff) with minimal time lag. This method is tested using a clinical dataset of ten real cases, each including glucose readings, insulin bolus, and meal (carbohydrate) data, as well as a dataset of ten simulated cases created from the UVA/Padova simulator. Four techniques are used to compare the performance of the recurrent Convolutional Neural Network. The suggested technique is implemented on an Android smartphone, with an execution time of 6 milliseconds on a phone against 780 milliseconds on a laptop.