Bidirectional cross-modal attention with adaptive gating for multimodal alzheimer's disease classification

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Imaging

Background:

• Alzheimer's disease (AD) is a progressive neurodegenerative disorder.
• Mild cognitive impairment (MCI) is a transitional stage between cognitively normal (CN) and AD.
• Early diagnosis of AD and MCI is crucial for timely intervention and disease management.

Purpose of the Study:

• To develop a multimodal deep learning framework for improved classification of AD and MCI.
• To overcome limitations of single-modal approaches and enhance cross-modal interaction modeling.
• To integrate structural magnetic resonance imaging (sMRI) and clinical features for a comprehensive diagnostic tool.

Main Methods:

• Utilized a 3D ResNet-34 for sMRI feature extraction and a multilayer perceptron (MLP) for clinical data encoding.
• Implemented a bidirectional cross-modal attention mechanism to strengthen feature associations.
• Employed an adaptive gated fusion module for dynamic integration of multimodal features.

Main Results:

• Achieved high accuracy in multi-class classification: 95.67% ± 1.70% (CN vs. MCI vs. AD) and 93.64% ± 2.36% (CN vs. EMCI vs. LMCI vs. AD).
• Demonstrated strong performance in binary classifications, including AD vs. CN (96.83% ± 1.53%) and MCI vs. CN (95.83% ± 1.39%).
• Results were validated for robustness using five different random seeds, reported as mean ± standard deviation.

Conclusions:

• The proposed multimodal deep learning framework offers an effective solution for computer-aided diagnosis of AD and MCI.
• Integrating sMRI and clinical data with advanced deep learning techniques significantly improves classification accuracy.
• This approach holds promise for early and accurate detection of cognitive impairment, aiding in disease progression management.