EXPLAINABLE AI IN THE CATEGORIZATION OF ALZHEIMER’S DISEASE

Abstract

Recent years have seen an unparalleled expansion in computing power, which has made it possible to create Artificial Intelligence (AI) models for medical applications with impressive outcomes. The usual blackbox nature of many AI models, how-ever, has hindered the acceptability and implementation of many AI-powered Computer Aided Diagnosis (CAD) techniques in the medical field. Thus, in order to encourage medical professionals to use these AI models, the algorithms’ predictions need to be comprehensible and interpretable. The goal of the new discipline of explainable AI (XAI) is to demonstrate why the predictions made by these models are reliable. The literature on Alzheimer’s disease (AD) detection with XAI that has been published in the past ten years is systematically reviewed in this paper. In order to classify AI models into various conceptual approaches (such as Post-hoc, Ante-hoc, Model-Agnostic, Model-Specific, Global, Local, etc.) and frameworks (such as Layer-wise Relevance Prop-agation, or LRP, Gradient-weighted Class Activation Mapping, or GradCAM, the Local Interpretable Model-Agnostic Explana-tion, or LIME, and SHapley Additive exPlanations, or SHAP), research questions were carefully formulated. This classification offers a wide range of interpretations, from intrinsic to global, by extending local explanations. Additionally, many interpretations that offer a thorough understanding of the elements supporting the clinical diagnosis of AD are also covered. Finally, XAI research’s needs, limitations, and unresolved issues are described, along with potential applications in AD detection