LEARNING TO SYNTHESIZE CORTICAL MORPHOLOGICAL CHANGES USING GRAPH CONDITIONAL VARIATIONAL AUTOENCODER.

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Medical Image Analysis

Background:

• Cortical thinning is a key indicator of brain aging and neurodegenerative diseases.
• Predicting individual brain morphology changes is crucial for early diagnosis and intervention.
• Existing methods often overlook the intricate surface topology of the brain.

Purpose of the Study:

• To develop a generative neural network model capable of predicting cortical thickness maps at future ages.
• To incorporate brain surface topology into the prediction model for improved accuracy.
• To enhance the understanding of temporospatial patterns in brain aging and neurodegeneration.

Main Methods:

• Utilized a conditional variational autoencoder (VAE) generative neural network, conditioned on age.
• Developed a novel loss function incorporating weighted adjacency to integrate mesh topology (edge connections) with cortical thickness (vertices).
• Compared the proposed model against traditional conditional VAEs lacking topological information.

Main Results:

• The proposed model demonstrated superior prediction of future cortical thickness maps compared to traditional VAEs.
• Prediction accuracy improved significantly, particularly with wider age gaps.
• The model successfully integrated surface topography into the age-conditional generative process.

Conclusions:

• The developed age-conditional VAE with topological integration offers a powerful tool for predicting individual cortical morphology.
• This approach enhances the understanding of aging-related brain changes and neurodegenerative disease trajectories.
• The model holds potential for personalized medicine and early disease detection.