ARTIFICIAL INTELLIGENCE IN DIABETES DETECTION: A COMPREHENSIVE REVIEW OF METHODS, CHALLENGES, AND FUTURE DIRECTIONS

Abstract

Background: Diabetes is a chronic metabolic condition that disturbs over 537 million adults global. Early and accurate detection is critical to prevent severe difficulties, containing cardiovascular illness, neuropathy, and retinopathy. Artificial intelligence (AI) has developed as a transformative approach for diabetes screening, leveraging machine learning, deep learning, and hybrid models to improve the detection.

Objective: This review synthesizes current AI methodologies for diabetes detection, evaluates their performance across diverse data sources, identifies key challenges, and explores innovative solutions to bridge the clinical implementation gaps.

Methods: A comprehensive analysis of AI-driven diabetes screening was conducted, focusing on methodologies applied to electronic health records (EHRs), medical imaging (e.g., retinal fundoscopy), and wearable-sensor data. The performance metrics, limitations, and clinical implications of these techniques were critically evaluated.

Results: AI models achieved high diagnostic accuracy (AUC: 0.82–0.95) across retrospective studies but exhibited performance degradation in real-world settings due to (i) data heterogeneity (up to 40% accuracy drop across healthcare systems) and (ii) algorithmic bias (sensitivity differences >15% across ethnic groups).

Conclusion: AI demonstrates strong potential for early diabetes detection but requires solutions for real-world applications. Prioritizing explainable frameworks, bias mitigation, and multicenter validation is critical for clinical adoption. More investigation is required to begin long-term effectiveness via different techniques.