Group-informed attentive framework for enhanced diabetes mellitus progression prediction

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Computational Biology

Background:

• Diabetes Mellitus (DM) is a growing global health issue requiring accurate progression prediction.
• Deep learning offers advanced predictive capabilities but faces challenges with incomplete data and model robustness.

Purpose of the Study:

• To improve the precision and reliability of diabetes progression predictions using deep learning.
• To address challenges of incomplete data and enhance predictive model robustness in DM research.

Main Methods:

• Utilized targeted imputation strategies for missing data within patient clusters.
• Implemented data augmentation and group-level feature analysis for model robustness.
• Developed a deep attentive transformer sensitive to group characteristics for processing diverse clinical data.

Main Results:

• The deep attentive transformer model accurately predicts DM progression using clinical and physical examination data.
• The model performs advanced feature selection and reasoning, identifying key individual and group-level factors.
• Demonstrated enhanced precision and reliability in diabetes progression prediction.

Conclusions:

• The developed deep learning framework significantly advances diabetes progression prediction.
• Provides deeper insights into factors influencing DM progression, aiding management and research.
• Highlights the potential of attentive transformers in handling complex healthcare data for chronic disease prediction.