An AI-based module for interstitial glucose forecasting enabling a "Do-It-Yourself" application for people with type...

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Endocrinology

Background:

• Diabetes mellitus (DM) affects over 500 million adults, with Type 1 diabetes (T1D) requiring insulin therapy and challenging glucose control.
• Current mHealth tools and deep learning (DL) models for glucose prediction often lack long-term user engagement, limiting their benefit in daily T1D self-management.
• Accurate short-term glucose level prediction is crucial for effective T1D management.

Purpose of the Study:

• To develop a do-it-yourself (DIY) deep learning (DL) framework for personalized interstitial glucose prediction using continuous glucose monitoring (CGM) data.
• To enable the generation of a unique DL model for each user, without relying on data from other individuals.
• To provide accurate glucose level predictions up to one hour ahead for improved T1D self-management.

Main Methods:

• A DIY module was created to process raw CGM data, preparing it for training and validation of a DL model.
• A personalized DL model was generated for each user, utilizing their own CGM data.
• One year of CGM data from 29 T1D subjects was used for model training and validation.

Main Results:

• The DL-based DIY framework demonstrated prediction performance comparable to state-of-the-art methods, using only CGM data.
• This work represents the first DL-based DIY approach for fully personalized glucose prediction.
• The open-source framework is deployable via Docker, allowing standalone use, smartphone integration, or further DL architecture experimentation.

Conclusions:

• The developed DIY framework offers a novel, personalized approach to glucose prediction for T1D management.
• The system's accuracy and personalization capabilities have the potential to enhance user engagement and improve T1D self-care.
• The open-source and adaptable nature of the framework facilitates broader adoption and future research in AI-driven diabetes management.