Alzheimer's disease prediction algorithm based on de-correlation constraint and multi-modal feature interaction.

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Biomedical Data Analysis

Background:

• Alzheimer's disease (AD) is a progressive neurodegenerative disorder.
• Multimodal brain imaging, including FDG-PET and MRI, offers complementary insights into AD pathology.
• Integrating multimodal data presents challenges due to differing feature spaces, potentially hindering prediction accuracy.

Purpose of the Study:

• To develop an advanced Alzheimer's disease prediction model.
• To effectively address the feature space discrepancies in multimodal brain imaging data.
• To enhance the utilization of complementary information between FDG-PET and MRI for improved AD prediction.

Main Methods:

• A novel AD prediction model incorporating de-correlation constraints and multi-modal feature interaction.
• Feature extraction using residual connections and attention mechanisms for FDG-PET and MRI data.
• A mutual attention feature fusion module designed to enhance inter-modal feature interaction and adaptive weighting.

Main Results:

• The proposed model achieved a prediction accuracy of 86.79% for distinguishing Alzheimer's disease (AD), Mild Cognitive Impairment (MCI), and Normal Cognition (NC).
• Demonstrated superior performance compared to existing multi-modal AD prediction models.
• The de-correlation constraint and mutual attention fusion effectively improved the model's ability to leverage complementary information from different imaging modalities.

Conclusions:

• The developed model effectively integrates FDG-PET and MRI data for accurate Alzheimer's disease prediction.
• The proposed de-correlation constraint and mutual attention fusion strategies are key to overcoming multimodal data challenges.
• This approach shows significant potential for improving early diagnosis and prediction of Alzheimer's disease.