Systematic review of explainable AI in Alzheimer’s diagnosis

Abstract

This study aims to provide a comprehensive and structured understanding of Explainable Artificial Intelligence (XAI) approaches used in the diagnosis of Alzheimer’s Disease (AD). It seeks to bridge the gap between emerging XAI techniques and their clinical applicability, addressing the urgent need for transparent and interpretable diagnostic tools. A systematic literature review was conducted using a structured search strategy to identify relevant studies published in the last five years. A total of 37 peer-reviewed articles were included, focusing on the application of XAI techniques—such as LIME, SHAP, Grad-CAM, and other emerging frameworks—within Machine Learning (ML) and Deep Learning (DL) models for AD diagnosis. The review reveals a growing interest in integrating XAI methods into clinical workflows, highlighting their potential to enhance diagnostic reliability and transparency. It presents a comparative analysis of major XAI frameworks, evaluating their effectiveness, interpretability, and suitability for clinical adoption. Key challenges identified include a lack of standardization across studies, limited dataset availability, and difficulties in generalizing findings. Several research gaps are noted, particularly in the consistency of XAI implementation and interpretability across different ML/DL models. XAI offers promising enhancements to AD diagnosis, but the field is still developing. Standardized methodologies, larger datasets, and improved generalization capabilities are essential for advancing clinical adoption. This review lays the groundwork for future research by identifying critical gaps and suggesting directions for the development of more interpretable and robust XAI models. The insights provided in this review can guide researchers, clinicians, and developers in selecting appropriate XAI frameworks for AD diagnosis. It also underscores the importance of interpretability in AI-driven healthcare applications, helping to foster trust and usability in real-world clinical settings.