Explainable Early Alzheimer's Disease Detection Using a Hybrid Capsule and Vision Transformer-Based Framework with LIME

Alzheimer's disease (AD) is a degenerative neurological disorder that results in important cognitive impairment. Early and accurate diagnosis is important for timely intervention and enhanced patient consequences. Traditional deep learning (DL) approaches employed in medical imaging, on the other hand, usually fail to capture hierarchical and spatial feature representations accurately. To address these limitations, this study introduces a new hybrid diagnostic framework known as the Multi-Level Capsule Network with Multi-Vision Transformer Attention (MCN-MVT) for early-stage Alzheimer's disease diagnosis utilizing neuroimaging data. The proposed model employs capsule networks to preserve spatial hierarchies and extract correct local information, while Vision Transformers (ViT) are employed to replicate long-distance dependencies and global contextual relationships among brain areas. Furthermore, Local Interpretable Model-Agnostic Explanations (LIME) progress model transparency and clinical interpretability by offering post-hoc explanations for predictions. The MCN-MVT framework was tested on a publicly available Alzheimer's MRI dataset and achieved 98.80% accuracy, 89.64% precision, 80.10% recall, and an AUC of 0.982. Comparative analysis reveals that the suggested technique outperforms cutting-edge models such as 3D-CNN, MCAD, and LDM-DDPM. These results display that the proposed technique is a reliable, interpretable, and high-performance solution for early Alzheimer's disease diagnosis, with promising real-world clinical applications.