Enhancing Personalized Treatment in Diabetes Using Genomic Data and Deep Learning Models: A Systematic Review

Abstract

Diabetes remains a significant global health challenge, necessitating innovative approaches for early detection and personalized treatment. Recent advancements in deep learning and genomic research have revolutionized diabetes prediction and management by enabling more accurate and individualized interventions. This systematic review explores the integration of convolutional neural networks (CNNs), recurrent neural networks (RNNs), and hybrid AI models in analysing genomic data for diabetes risk prediction, complication forecasting, and treatment optimization. The findings highlight that hybrid deep learning models outperform traditional machine learning techniques, achieving a predictive accuracy of up to 94.6%. Statistical validation using the Wilcoxon signed-rank and Kruskal-Wallis tests confirm the robustness of these models across diverse genomic datasets. Explainable AI techniques such as SHAP and LIME have enhanced model transparency, improving clinician trust by 45%. Additionally, federated learning frameworks have demonstrated a 30% improvement in scalability, ensuring privacy-preserving genomic analysis. Despite these advancements, challenges remain in data heterogeneity, model interpretability, and computational efficiency. Addressing these limitations through improved statistical validation, standardized genomic data integration, and explainable AI frameworks will be crucial for the widespread clinical adoption of AI-driven personalized diabetes care.