Advanced AI Techniques for Real-Time Blood GlucosePrediction in Diabetics: A Study Using Deep Learningand Genetic Algorithms

Abstract

Diabetes mellitus is one of the most common chronic diseases
worldwide, with its prevalence expected to rise sharply. Projections
suggest that by 2050, more than 1.3 billion people globally will be
living with diabetes, a significant increase from the current estimate
of 529 million. A case of Type I diabetes is characterized by the
pancreas failing to produce adequate amounts of insulin, which leads
to uncontrolled blood glucose levels. Traditionally, management
involves patient-administered insulin and monitoring blood glucose
levels (BGLs) based on dietary intake reported by the patient.
 This study introduces an innovative method that leverages advanced
Artificial Intelligence (AI) techniques to continuously predict blood
glucose levels for the short term (+30 minutes) from the current
situation. The techniques applied include Deep Learning with Artificial
Neural Networks (ANNs), Genetic Algorithms (GAs), and Reinforcement
Learning. These methods analyzed both raw BGL data and additional
information derived from a Diabetic Dynamic Model of BGLs.
 The study’s preliminary evaluation used data from four virtual patients
generated by an open-source diabetes simulation tool and three
real diabetic patients using the DexCom SEVEN system. The results
indicated that the knowledge-based approach significantly enhanced
prediction accuracy, with Genetic Algorithms outperforming ANNs.
Additionally, the integration of online learning and Reinforcement
Learning, which adapt to emerging data patterns, further improved
predictive accuracy.
 This advanced methodology demonstrates considerable potential
for enhancing diabetes management by providing timely and precise
BGL predictions without direct patient input. Future studies involving
larger cohorts of both Type I and Type II diabetic patients are necessary
to validate these promising results.