Convolutional Recurrent Neural Networks for Glucose Prediction (1807.03043v5)

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$\pm$0.71 [mg/dL] over a 30-minute horizon, RMSE = 18.87$\pm$2.25 [mg/dL] over a 60-minute horizon) and real patient cases (RMSE = 21.07$\pm$2.35 [mg/dL] for 30-minute, RMSE = 33.27$\pm$4.79\% for 60-minute). In addition, the model provides competitive performance in providing effective prediction horizon ($PH_{eff}$) with minimal time lag both in a simulated patient dataset ($PH_{eff}$ = 29.0$\pm$0.7 for 30-min and $PH_{eff}$ = 49.8$\pm$2.9 for 60-min) and in a real patient dataset ($PH_{eff}$ = 19.3$\pm$3.1 for 30-min and $PH_{eff}$ = 29.3$\pm$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 $6$ms on a phone compared to an execution time of $780$ms on a laptop.

Overview of "Convolutional Recurrent Neural Networks for Glucose Prediction"

The paper "Convolutional Recurrent Neural Networks for Glucose Prediction" by Kezhi Li et al. explores the application of deep learning techniques to predict blood glucose levels for Type 1 diabetes mellitus (T1DM) management. This paper aims to enhance diabetes treatment by accurately predicting glucose levels, thus allowing for timely adjustments and interventions to prevent hypo- or hyperglycemic events. The authors present a novel approach utilizing Convolutional Recurrent Neural Networks (CRNNs) to improve glucose prediction accuracy over traditional baselines.

Summary of Methods

The proposed CRNN model leverages a combination of Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTM) networks to capture both spatial and temporal patterns in glucose data. The architecture is designed to forecast glucose levels by considering various inputs such as continuous glucose monitoring (CGM) data, insulin delivery, and carbohydrate intake:

• Data Sources: The model was evaluated on data generated from the UVa/Padova simulator for simulated cases and a clinical dataset from real patient cases. Both datasets include glucose readings, insulin doses, and meal information.
• Model Architecture: The CRNN architecture integrates multiple layers:
  • Convolutional Layers: To extract features from input data such as glucose, carbohydrate, and insulin time series.
  • LSTM Layers: To learn long-term dependencies in the sequence data, critical for the dynamic and variable nature of glucose levels.
  • Fully-Connected Layers: To produce the final output, which is the predicted glucose level.
• Training and Implementation: The model is trained using TensorFlow and further optimized for deployment on mobile devices via TensorFlow Lite, achieving execution times as low as 6ms on mobile phones.

Results

The CRNN model demonstrated superior performance in glucose prediction when compared with traditional methods including Support Vector Regression (SVR), autoregressive models (ARX), and neural network predictive algorithms (NNPG):

• In Silico Performance: The model achieved a root mean square error (RMSE) of 9.38 mg/dL over a 30-minute prediction horizon and 18.87 mg/dL over a 60-minute horizon for simulated data, significantly outperforming benchmark methods.
• Clinical Dataset Performance: On real patient data, the model recorded an RMSE of 21.07 mg/dL for 30 minutes and 33.27 mg/dL for 60 minutes. These results reflect a considerable improvement over baseline approaches, particularly in providing effective prediction horizons and detecting adverse glycemic events.

Implications and Future Prospects

The integration of CNN and LSTM components allows the CRNN model to effectively capture intricate patterns in glucose dynamics, contributing to its high accuracy. The practical implications of this work are substantial; by enabling anticipatory glycemic management through mobile platforms, the model enhances the potential for personalized diabetes care.

Theoretically, this approach pushes the envelope in medical AI, illustrating the robustness of deep learning models for real-time biomedical applications. Moving forward, incorporating additional contextual data like stress levels or physical activity could potentially bolster prediction accuracy further. Continuous model refinement and validation on more extensive clinical datasets will be crucial for widespread adoption.

In conclusion, the paper establishes a compelling case for the application of deep learning in glucose monitoring, setting the groundwork for future research and development in AI-driven health management systems.