Brain Latent Progression: Individual-based spatiotemporal disease progression on 3D Brain MRIs via latent diffusion

Area of Science:

• Artificial Intelligence
• Neuroimaging
• Medical Image Analysis

Background:

• Longitudinal MRI datasets enable AI-driven disease progression modeling.
• Current AI methods struggle with patient-specific prediction, spatiotemporal consistency, and memory demands for 3D brain MRIs.

Purpose of the Study:

• To propose Brain Latent Progression (BrLP), a novel spatiotemporal model for predicting individual disease progression in 3D brain MRIs.
• To address limitations in patient individualization, spatiotemporal consistency, longitudinal data utilization, and computational efficiency.

Main Methods:

• BrLP operates in a reduced latent space to manage high-dimensional data.
• Integrates subject metadata for enhanced individualization and prior disease dynamics knowledge via an auxiliary model.
• Employs the Latent Average Stabilization (LAS) algorithm for spatiotemporal consistency and uncertainty quantification.

Main Results:

• BrLP was trained and evaluated on over 11,730 T1-weighted brain MRIs from 2,805 subjects.
• Validated on an external dataset of 2,257 MRIs from 962 subjects.
• BrLP demonstrated state-of-the-art accuracy in predicting follow-up MRIs compared to existing methods.

Conclusions:

• BrLP offers a robust solution for predicting individual brain disease progression from longitudinal MRI data.
• The model achieves high accuracy, ensures spatiotemporal consistency, and quantifies prediction uncertainty.
• Publicly available code facilitates further research and application in medical image analysis.