A Multimodal Explainable AI Framework for Early Detection of Alzheimer's Disease Using MRI, PET, and Clinical...

Early and accurate detection of Alzheimer's disease (AD), especially at the Mild Cognitive Impairment (MCI) stage, remains a critical challenge in clinical neurology. While deep learning models have achieved notable success in diagnosing AD using neuroimaging and clinical data, their black-box nature hin- ders trust and adoption in medical settings. This paper proposes a novel multimodal deep learning framework that integrates structural MRI, PET scans, and clinical assessments such as MMSE, APOE status, and age for AD classification. In order to combine modality- specific representations, a transformer- based fusion model is created. To improve model transparency, explainable AI (XAI) techniques are used, such as Grad-CAM for imaging data and SHAP for clinical features. Comparing the ADNI dataset to current unimodal or inexplicable models, experiments show notable gains in classification accuracy and interpretability. The suggested framework is a step toward reliable AI systems in neurology since it offers both strong predictions and therapeutically applicable insights.