AD-Diff: enhancing Alzheimer's disease prediction accuracy through multimodal fusion

Area of Science:

• Neuroimaging
• Artificial Intelligence in Medicine
• Biomedical Data Analysis

Background:

• Early Alzheimer's disease (AD) detection is vital for patient care and treatment efficacy.
• Current predictive models struggle with integrating multimodal data and the high cost of PET imaging.

Purpose of the Study:

• To develop an innovative model, AD-Diff, for improved early prediction of Alzheimer's disease.
• To overcome limitations in multimodal data integration and PET image acquisition costs.

Main Methods:

• The AD-Diff model integrates PET images generated via a 3D diffusion process with cognitive scale data and MRI.
• A novel ADdiffusion module generates high-quality PET images.
• A multimodal Mamba Classifier processes fused imaging and tabular data.

Main Results:

• AD-Diff demonstrated exceptional performance in both long-term and short-term AD prediction tasks on OASIS and ADNI datasets.
• The model significantly improved prediction accuracy and reliability compared to existing methods.
• Validation on OASIS and ADNI datasets confirmed model efficacy.

Conclusions:

• The AD-Diff model offers a powerful approach for early Alzheimer's disease diagnosis by effectively integrating multimodal data.
• This innovative method addresses key challenges in current predictive strategies, paving the way for personalized treatment.