Forecasting Alzheimer's disease progression via identity-preserved denoising diffusion generative adversarial network

Area of Science:

• Medical imaging
• Artificial intelligence
• Neuroscience

Background:

• Accurate forecasting of Alzheimer's disease (AD) progression is crucial for assessing preventative interventions.
• Predicting longitudinal MRI changes in AD is challenging, especially maintaining subject identity with generative models.

Purpose of the Study:

• To develop a novel deep learning model for generating subject-specific longitudinal MRIs.
• To ensure the generated MRIs preserve individual subject identity and biological accuracy.

Main Methods:

• Developed an identity-preserved denoising diffusion generative adversarial network (IP-DDGAN).
• Incorporated a metadata-guided module and regularization terms for identity preservation.
• Utilized morphometric, identity-consistency, and image-quality metrics for evaluation.

Main Results:

• IP-DDGAN successfully generated subject-specific longitudinal MRIs.
• Synthetic MRIs retained biological and disease-related phenotypes.
• The model accurately captured temporal changes and predicted distinct disease trajectories (CN to MCI, MCI to AD).

Conclusions:

• The proposed IP-DDGAN model can generate realistic and biologically plausible longitudinal MRIs.
• This method aids in predicting individual Alzheimer's disease progression trajectories.
• The model supports downstream applications for evaluating AD prevention strategies.