Enhancing Alzheimer’s Disease Diagnosis using Transparent AI Models: A Survey

The integration of artificial intelligence (AI) in Alzheimer’s disease detection has improved diagnostic capabilities. It enables accurate and early identification of the disease through advanced analysis of neuroimaging data, biomarkers, and cognitive assessments. The complexity of modern AI models is high and lack clinical adoption due to lack of transparency and interpretability. It creates a barrier for healthcare professionals to understand and trust the AI's decision-making process. This paper presents role of explainable artificial intelligence (XAI) to highlight the importance of high-performing AI models. The various XAI techniques are categorized based on their scope (global vs. local), implementation timing (ante-hoc vs. post-hoc), and applications (model-specific vs. model-agnostic). The global explanations provide an overall understanding of the model's behavior while local explanations focus on individual predictions. The ante-hoc models are interpretable from the start, and post-hoc models require additional techniques to explain their outputs. Also, model-specific approaches are used for particular algorithms, model-agnostic approaches are applied across different model types. The different XAI frameworks such as SHAP, LIME, Grad-CAM, and LRP are studied for AD classification to obtain model decisions without compromising accuracy. This paper also presents different challenges such as data imbalance and over fitting that provides a balance between model complexity and interpretability. The availability of small dataset of different stages of Alzheimer introduces scenes in the model’s performance. Also, model performs well during training and generates poor results during testing. It is observed that XAI algorithms have improved the transparency and trustworthiness in AI-assisted diagnostic systems for effective Alzheimer’s disease detection.