Convolutional Recurrent Neural Networks for Glucose Prediction

Abstract

Control of blood glucose is essential for diabetes management. Current digital therapeutic approaches for subjects with Type 1 diabetes mellitus (T1DM) such as the artificial pancreas and insulin bolus calculators leverage machine learning techniques for predicting subcutaneous glucose for improved control. Deep learning has recently been applied in healthcare and medical research to achieve state-of-the-art results in a range of tasks including disease diagnosis, and patient state prediction among others. In this work, we present a deep learning model that is capable of forecasting glucose levels with leading accuracy for simulated patient cases (RMSE = 9.38±0.71 [mg/dL] over a 30-minute horizon, RMSE = 18.87±2.25 [mg/dL] over a 60minute horizon) and real patient cases (RMSE = 21.07±2.35 [mg/dL] for 30-minute, RMSE = 33.27±4.79% for 60-minute).In addition, the model provides competitive performance in providing effective prediction horizon (P H ef f ) with minimal time lag both in a simulated patient dataset (P H ef f = 29.0±0.7 for 30-min and P H ef f = 49.8±2.9 for 60-min) and in a real patient dataset (P H ef f = 19.3±3.1 for 30-min and P H ef f = 29.3±9.4 for 60-min). This approach is evaluated on a dataset of 10 simulated cases generated from the UVa/Padova simulator and a clinical dataset of 10 real cases each containing glucose readings, insulin bolus, and meal (carbohydrate) data. Performance of the recurrent convolutional neural network is benchmarked against four algorithms. The proposed algorithm is implemented on an Android mobile phone, with an execution time of 6ms on a phone compared to an execution time of 780ms on a laptop.

Dilated Recurrent Neural Networks for Glucose Forecasting in Type 1 Diabetes

“…Using RMSE and MARD to evaluate the glucose prediction is a common approach in many previous works [6,9,22,30]. RMSE can effectively present the overall performance of the prediction models.…”

“…MARD reflects the relative error to the current glucose levels and alters the risk of hypoglycemic [31]. The time lag evaluates how fast the prediction models react to the abrupt changes of BG levels [9,30].…”

“…For example, we can use shallow neural networks as the bottom layers, such as fully connected or convolutional layers to extract features and pre-process the data. Then, the feature maps are fed into the DRNN to obtain high-level features of time series, similar to [30]. For state-of-the-art techniques, we consider using the DRNN layer as a basic unit in generative adversarial network and reinforcement learning for diabetes management.…”

Predicting Quality of Overnight Glycaemic Control in Type 1 Diabetes Using Binary Classifiers

“…Machine learning techniques are becoming increasingly popular to solve many T1D management problems, such as glucose forecasting [19], optimal insulin dosing [9,10], patient risk stratification [20], and CGM fault detection [21], as a result of their ability to represent complex non-linear input-output relationships, such as glucose-insulin dynamics. They can also be used to classify the quality of glycaemic control.…”

Deep Learning on Computational-Resource-Limited Platforms: A Survey

“…By contrast with traditional machine learning solutions, deep learning techniques are undergoing rapid development. Applications of deep learning involve information retrieval [4], natural language processing [5], human voice recognition [6], computer vision [7], anomaly detection [8], recommendation systems [9], bioinformatics [10], medicine [11,12], crop science [13], earth science [14], robotics [15][16][17][18], transportation engineering [19], communication technologies [20][21][22], and system simulation [23,24].…”