An intelligent diabetes software prototype: predicting blood glucose levels and recommending regimen changes

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Diabetes Management

Background:

• Maintaining optimal blood glucose (BG) control is challenging for type 1 diabetes mellitus (T1DM) patients, especially when daily regimens of food, insulin, and exercise are altered.
• Artificial intelligence (AI) systems, utilizing treatment algorithms calibrated with patient-specific data, may offer a solution for predicting and compensating for BG level changes due to regimen disturbances.

Purpose of the Study:

• To develop and evaluate a software prototype for a patient-specific BG prediction model using AI concepts.
• To determine the feasibility and efficacy of AI in assisting T1DM patients with BG management.

Main Methods:

• A software prototype was developed incorporating neural network, fuzzy logic, and expert system concepts.
• The model was calibrated using patient-specific BG data, including daily insulin, food, and exercise information.
• Evaluation involved calculating the Mean Absolute Percent Error (MAPE) between actual and predicted BG values.

Main Results:

• The calibrated AI model achieved a Mean Absolute Percent Error (MAPE) of 10.5% for BG prediction in a T1DM test subject.
• The prototype demonstrated viability as a learning tool for diabetes patients, although its predictive accuracy was limited by factors like human error and situational circumstances.

Conclusions:

• The AI prototype shows potential for personalized diabetes management by predicting BG fluctuations.
• Further development is warranted, especially with advancements in diagnostic and data capture tools to improve model accuracy and reduce testing requirements.
• The study highlights the need for more extensive evaluation with larger sample sizes to establish model validity.