A Survey of Methods and Architectures for Blockchain-Based Hybrid Contextual-ATNet Approach for Daily Diabetes Management: Predicting Insulin Dosage for Improved Control

Abstract

Diabetes mellitus is a major global health challenge requiring precise and personalized insulin dosage management to prevent severe complications. Traditional approaches, including rule-based dosing and periodic clinical adjustments, are inadequate for handling the complex, dynamic interactions between physiological, behavioral, and environmental factors influencing blood glucose levels. The availability of continuous glucose monitoring data has created opportunities for advanced computational systems to improve glycemic control.  This survey reviews artificial intelligence-based approaches for insulin dosage prediction, with a focus on blockchain-integrated Hybrid Contextual Attention Networks (ATNet). The ATNet framework combines attention-based temporal modeling with contextual feature fusion to process multimodal inputs such as glucose readings, dietary patterns, physical activity, and stress indicators. This architecture enables accurate prediction of glucose trends and personalized insulin recommendations by capturing complex temporal and contextual dependencies. Blockchain integration enhances the system by providing secure, decentralized data management, enabling privacy-preserving federated learning, and ensuring auditability through smart contracts. Applications include real-time glucose prediction, personalized insulin therapy, and clinical decision support. Comparative analysis demonstrates improved predictive accuracy and robustness over traditional methods. However, challenges such as scalability, interpretability, and clinical validation remain, highlighting the need for further research in developing reliable and deployable AI-driven diabetes management systems.