Feature-Level Fusion of T2- and PD-Weighted MRI Using Vision Transformer for Five-Stage Alzheimer’s Disease Classification

Abstract

Alzheimer’s disease (AD) is a progressive neurodegenerative disorder in which early and accurate diagnosis remains challenging, particularly for clinically overlapping mild cognitive impairment (MCI) stages. Although magnetic resonance imaging (MRI) is widely used for AD assessment, reliance on a single imaging sequence often limits discrimination between disease stages. Feature fusion has recently emerged as an effective strategy to enhance classification performance while maintaining clinical feasibility. In this study, an efficient feature-level fusion framework is proposed that integrates complementary information from T2-weighted and proton density (PD-weighted) MRI. Deep features are extracted exclusively using a Vision Transformer (ViT) model and combined through a weighted fusion strategy, followed by k-nearest neighbors classification. Experiments conducted on the ADNI dataset demonstrate that the proposed approach achieves high discriminative performance across five classes (AD, CN, EMCI, LMCI, and MCI), reaching an overall accuracy of approximately 99.35%. The fusion framework consistently outperforms single-modality models, particularly in distinguishing transitional stages. These results indicate that ViT-based feature fusion of T2 and PD MRI provides a robust and computationally efficient solution for multi-stage Alzheimer’s disease classification, with strong potential for practical clinical application.