Explainable AI predicting Alzheimer's disease with latent multimodal deep neural networks.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Informatics

Background:

• Alzheimer's disease (AD) is a progressive neurodegenerative disorder causing cognitive decline.
• Accurate prediction of AD cognitive status is crucial for timely intervention and management.

Purpose of the Study:

• To develop and evaluate a novel latent multimodal deep learning framework for predicting Alzheimer's disease cognitive status.
• To assess the contribution of clinical, neuroimaging, and genetic data in predicting AD cognitive impairment.

Main Methods:

• Utilized data from 322 patients (aged 55-92) from the ADNI database.
• Applied Confirmatory Factor Analysis (CFA) to derive latent cognitive impairment scores.
• Constructed a multimodal deep neural network with attention and cross-attention layers, integrating clinical, imaging, and genetic data.

Main Results:

• The multimodal neural network incorporating clinical and imaging data with attention layers achieved the best predictive performance (MAE: 0.330, MSE: 0.206).
• Clinical data was identified as the most influential modality, contributing 35% to the prediction of AD cognitive status.
• Attention mechanisms significantly enhanced the model's predictive capabilities.

Conclusions:

• The attention-based multimodal deep learning model demonstrates superior performance in predicting Alzheimer's disease cognitive impairment.
• Integrating multiple data modalities, particularly clinical and neuroimaging data, is effective for AD cognitive status prediction.
• Attention layers are a valuable addition to deep learning models for improving prediction accuracy in neurodegenerative disease research.