A U-Net-based multimodal deep learning model for high-precision blood glucose prediction using non-invasive...

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Metabolic Disease Research

Background:

• Accurate blood glucose prediction is crucial for managing Type 1 Diabetes (T1DM).
• Current methods often rely on invasive glucose monitoring.
• Integrating non-invasive physiological data offers a promising alternative for continuous monitoring.

Purpose of the Study:

• To develop and validate a novel U-Net-based convolutional neural network (CNN) for short-term blood glucose prediction.
• To integrate historical continuous glucose monitoring (CGM) data with non-invasive physiological parameters.
• To evaluate the model's performance and non-invasiveness for potential wearable sensor applications.

Main Methods:

• A U-Net-based CNN architecture was employed.
• The model utilized historical CGM data and non-invasive physiological parameters.
• The OhioT1DM dataset was used for training and validation.

Main Results:

• The model demonstrated robust performance on the OhioT1DM dataset.
• Achieved Mean Absolute Errors (MAEs) of 8.4761 mg/dL (30 min) and 14.0170 mg/dL (60 min).
• Reported Root Mean Square Errors (RMSEs) of 13.1315 mg/dL and 20.6470 mg/dL, with R-squared values of 0.9397 and 0.8549, respectively.

Conclusions:

• The proposed U-Net framework provides a practical, data-driven approach for short-term glucose forecasting in T1DM patients.
• The model's performance indicates its potential for improving metabolic health management.
• Quantitative validation of non-invasiveness supports its use in developing wearable glucose monitoring systems.