Blockchain-Based Hybrid ATNet for Intelligent Daily Diabetes Insulin Prediction

Abstract

Diabetes mellitus is a widespread chronic disease requiring precise insulin dosage management to prevent severe complications such as hyperglycemia and hypoglycemia. Traditional manual methods often fail to address the dynamic and individualized nature of glucose regulation influenced by diet, activity, and physiological variability. This has driven the need for intelligent, data-driven systems capable of improving personalized diabetes care.

This paper presents a comprehensive review of a Blockchain-Based Hybrid Contextual Attention Network (ATNet) for insulin dosage prediction. The model integrates attention-based temporal learning with contextual feature extraction, enabling it to capture complex relationships from continuous glucose monitoring, dietary intake, physical activity, and emotional factors. By assigning importance to relevant temporal patterns, ATNet improves prediction accuracy compared to conventional deep learning models.

Blockchain integration enhances the framework by providing secure, decentralized data management through smart contracts, ensuring privacy, auditability, and federated learning capabilities. Applications include real-time glucose prediction, personalized insulin recommendations, and clinical decision support. Empirical results demonstrate improved accuracy and reliability across benchmark datasets. However, challenges such as scalability, interoperability, and clinical deployment persist. This review highlights the potential of combining deep learning and blockchain to develop secure, efficient, and patient-centric diabetes management systems.