Identifying Progression-Specific Alzheimer's Subtypes Using Multimodal Transformer.

Area of Science:

• Neurodegenerative Diseases
• Biomedical Data Science
• Computational Neuroscience

Background:

• Alzheimer's disease (AD) is the most common neurodegenerative disorder, but current treatments are limited and their efficacy is uncertain due to disease heterogeneity.
• Early identification of AD subtypes is essential for effective intervention, but predicting these in asymptomatic or prodromal stages is challenging.
• Existing classification models often lack explainability and rely on single data modalities, limiting their predictive power.

Purpose of the Study:

• To develop and validate a multimodal framework for early classification of Alzheimer's disease progression subtypes.
• To introduce a tri-modal co-attention mechanism (Tri-COAT) for capturing cross-modal feature associations.
• To enhance the explainability of AD subtype classification models.

Main Methods:

• A multimodal framework integrating neuroimaging, genetic, and clinical assessment data was developed.
• A novel tri-modal co-attention mechanism (Tri-COAT) was introduced to model inter-modal feature dependencies.
• The framework was trained and evaluated on data from the Alzheimer's Disease Neuroimaging Initiative (ADNI) using 10-fold cross-validation.

Main Results:

• The Tri-COAT framework demonstrated superior performance compared to baseline models in classifying AD progression subtypes.
• The model achieved the highest classification area under the receiver operating characteristic curve.
• The co-attention mechanism provided interpretability by highlighting essential cross-modal feature associations.

Conclusions:

• The proposed multimodal framework effectively classifies Alzheimer's disease progression subtypes at early stages.
• Tri-COAT offers an interpretable approach by leveraging cross-modal associations from imaging, genetics, and clinical data.
• This methodology holds promise for advancing personalized medicine in Alzheimer's disease research.