Learning lifespan brain anatomical correspondence via cortical developmental continuity transfer

Area of Science:

• Neuroscience
• Computer Vision
• Medical Imaging

Background:

• Accurate anatomical correspondence is crucial for understanding human brain development and aging.
• Individual variability in cortical folding and limited neuroimaging data pose challenges for fine-scale lifespan anatomical mapping.

Purpose of the Study:

• To develop a novel transfer learning strategy for robustly inferring lifespan anatomical correspondences in the human brain.
• To address the limitations of scarce data in specific neurodevelopmental stages.

Main Methods:

• A transfer learning framework leveraging the developmental continuity of the cerebral cortex.
• Training a model on a large dataset and adapting it to other age groups along the developmental trajectory.
• Utilizing a novel loss function to preserve common patterns and capture group-specific features during transfer.

Main Results:

• The proposed transfer learning strategy significantly enhances model performance for populations with limited training samples, such as during early neurodevelopment.
• The framework enables robust inference of complex many-to-many anatomical correspondences across different neurodevelopmental stages.
• Evaluation involved multiple datasets encompassing over 1,000 brains from 34 weeks gestation to young adulthood.

Conclusions:

• The novel transfer learning approach effectively overcomes data scarcity issues in neurodevelopmental studies.
• This method provides a robust solution for mapping brain anatomical correspondences throughout the human lifespan.
• The findings facilitate more accurate studies of brain development and aging.