A multi-view multimodal deep learning framework for Alzheimer's disease diagnosis.

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Medical Diagnostics

Background:

• Early Alzheimer's disease (AD) diagnosis is challenging due to similarities with mild cognitive impairment (MCI) and cognitively normal (CN) individuals.
• Population heterogeneity, label noise, and imaging variations further complicate diagnosis.
• Current multimodal neuroimaging approaches have limitations in fusing complementary data and aggregating multi-scale features.

Purpose of the Study:

• To develop a novel multimodal diagnostic framework, ADMV-Net, for enhanced recognition accuracy across all Alzheimer's disease stages.
• To improve the fusion of multimodal neuroimaging data and multi-scale feature extraction for more robust AD diagnosis.

Main Methods:

• Proposed ADMV-Net framework utilizes a dual-pathway Hybrid Convolution ResNet for 3D medical image feature extraction, fusing global and local information.
• Implemented a Multi-view Fusion Learning mechanism with Global Perception, Multi-level Local Cross-modal Aggregation, and Bidirectional Cross-Attention modules.
• Incorporated a Regional Interest Perception Module to focus on AD-pathology-associated brain regions.

Main Results:

• ADMV-Net achieved 94.83% accuracy and 95.97% AUC in distinguishing Alzheimer's disease (AD) from cognitively normal (CN) individuals.
• The framework demonstrated superior performance compared to mainstream methods in AD vs. CN classification.
• Showcased strong discriminative capability and excellent generalization in multi-class classification tasks.

Conclusions:

• ADMV-Net effectively leverages multimodal and multi-view information to significantly improve Alzheimer's disease diagnostic accuracy.
• The framework's integration of global, local, and regional features offers a promising tool for early AD diagnosis and clinical decision-making.
• The developed framework provides a valuable resource for advancing Alzheimer's disease research and clinical practice.