GlucoNet-MM: A multimodal attention-based multi-task learning framework with decision transformer for personalised...

Area of Science:

• * Computational biology and bioinformatics
• * Artificial intelligence in healthcare
• * Diabetes mellitus research

Background:

• * Accurate blood glucose prediction is crucial for effective diabetes management.
• * Conventional machine learning models face challenges in patient-specific forecasting due to data variability and complexity.
• * Need for interpretable and adaptive AI solutions for personalized glycemic control.

Purpose of the Study:

• * To develop an interpretable deep learning (DL) framework for patient-specific, policy-aware blood glucose forecasting.
• * To integrate multimodal physiological and behavioral data for enhanced prediction accuracy.
• * To improve clinical trust and enable proactive diabetes management through advanced AI.

Main Methods:

• * Proposed GlucoNet-MM, a multimodal DL framework combining attention-based multi-task learning (MTL) and a Decision Transformer (DT).
• * Integrated heterogeneous data: continuous glucose monitoring (CGM), insulin dosage, carbohydrate intake, and physical activity.
• * Employed temporal attention, integrated gradients for interpretability, and Monte Carlo dropout for uncertainty quantification.

Main Results:

• * Achieved high performance on public datasets (BrisT1D, OhioT1DM) with R² scores of 0.94-0.96 and low MAE (0.027-0.031).
• * Demonstrated superior predictive accuracy and generalizability compared to single-modality and conventional baseline models.
• * Validated the model's effectiveness in capturing complex temporal dependencies in glycemic dynamics.

Conclusions:

• * GlucoNet-MM offers a significant advancement in intelligent, personalized clinical decision support for diabetes care.
• * The multimodal design, policy-aware forecasting, and interpretability enhance prediction accuracy and clinical trust.
• * Enables proactive glycemic management tailored to individual patient needs, improving diabetes outcomes.