AlzFormer: Multi-modal framework for Alzheimer's classification using MRI and graph-embedded demographics guided by...

Area of Science:

• Neuroscience
• Medical Imaging
• Artificial Intelligence

Background:

• Alzheimer's disease (AD) is a leading cause of death, posing significant diagnostic challenges due to subtle brain changes and limited integration of demographic data.
• Current machine learning models struggle with diffuse pathology, non-uniform MRI alterations, and demographic context, hindering accurate AD classification.
• Early and accurate detection of AD is crucial for effective patient management and treatment.

Purpose of the Study:

• To develop a novel multi-modal deep learning framework, AlzFormer, for enhanced Alzheimer's disease classification.
• To address limitations in capturing global pathology, processing multi-planar MRI data, and integrating demographic information in AD detection.
• To improve the accuracy and robustness of automated AD diagnosis by combining 3D MRI and demographic features.

Main Methods:

• Proposed a multi-modal deep learning framework, AlzFormer, integrating 3D Convolutional Neural Networks (CNNs) for volumetric features and parallel 2D CNNs with a Transformer encoder for tri-planar MRI analysis.
• Incorporated demographic features as knowledge graph embeddings using a novel Adaptive Attention Gating mechanism to dynamically balance MRI and demographic data contributions.
• Conducted comprehensive experiments on two real-world datasets, including generalization, ablation, and robustness tests.

Main Results:

• The AlzFormer model demonstrated robust and effective performance in Alzheimer's disease classification across multiple datasets.
• Ablation studies and robustness evaluations confirmed the model's effectiveness under noisy conditions and its ability to generalize.
• The proposed framework successfully integrated 3D MRI and demographic data, outperforming existing methods by addressing key limitations.

Conclusions:

• AlzFormer offers a powerful and interpretable solution for Alzheimer's disease diagnosis, significantly enhancing diagnostic accuracy.
• The model's ability to integrate multi-modal data suggests strong potential for integration into Clinical Decision Support Systems (CDSS).
• This approach paves the way for more personalized and accurate early detection of Alzheimer's disease.