Explainable AI for Detecting Early-Stage Alzheimer's Disease Using Multimodal Neuroimaging and Genomic Data

Alzheimer's disease (AD), a progressive neurodegenerative disorder, demands early and accurate diagnosis to enable timely intervention. Current artificial intelligence (AI) models for AD detection often lack interpretability, limiting their clinical adoption. This study introduces an explainable AI (XAI) framework that synergistically combines multimodal neuroimaging and genomic data to enhance early-stage AD diagnosis while demystifying model decisions. Leveraging the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset, we integrated structural MRI-derived hippocampal volumetry with APOE- $\varepsilon 4$ genotypic data from 1,002 participants ($\text{3 0 0}$ cognitively normal, 352 mild cognitive impairment [MCI], 350 AD). A hybrid architecture was developed, comprising a 3D convolutional neural network (CNN) for automated hippocampal segmentation, a gradient-boosting model (XGBoost) for genomic risk stratification, and an attention-based fusion layer to combine imaging and genomic embeddings. Post-hoc interpretability was achieved using SHapley Additive exPlanations (SHAP) and layer-wise relevance propagation (LRP). Our multimodal model achieved an area under the curve(AUC) of 0.92 (95 % CI: 0.89-0.95), significantly outperforming MRI-only (AUC: 0.85, $\mathrm{p}<0.001$) and genomic-only (AUC: 0.76, $\mathrm{p}<0.001$) baselines. Explainability analysis revealed that hippocampal atrophy contributed 58 % to AD predictions, while APOE- $\varepsilon 4$ carrier status amplified risk scores by $3.2 \times$ in patients with moderate neurodegeneration. These findings align with established AD pathophysiology, underscoring the framework's biological plausibility. By translating complex AI decisions into clinician-friendly explanations, this work bridges the gap between computational innovation and clinical trust. The proposed approach not only improves diagnostic accuracy but also identifies high-risk MCI patients for personalized therapeutic strategies. This study advances the translational potential of AI in neurodegenerative diseases, emphasizing transparency, multimodal integration, and reproducibility through open-source data.