Prediction of Longitudinal Development of Infant Cortical Surface Shape Using a 4D Current-Based Learning Framework

Area of Science:

• Neuroscience
• Medical Imaging
• Computational Biology

Background:

• Understanding early human brain development, particularly the cerebral cortex, is crucial but challenging.
• Longitudinal neuroimaging and modeling are advancing research but face limitations like sparse data and lack of subject matching.
• The dynamic evolution of infant cortical shape remains an area with many unanswered questions.

Purpose of the Study:

• To develop a novel framework for predicting infant cortical surface shape dynamics from a single baseline scan.
• To address challenges in longitudinal neuroimaging, including limited timepoints and inter-subject variability.
• To enable accurate prediction of cortical shape evolution throughout infancy.

Main Methods:

• A spatiotemporal (4D) current-based learning approach was developed for predicting dynamic shape evolution.
• The framework involves a training stage to learn geometric and dynamic features and establish correspondences.
• A prediction stage uses learned features to forecast cortical shapes at multiple future timepoints from a baseline scan.

Main Results:

• The proposed framework successfully predicted the inner cortical surface shape at 3, 6, and 9 months from birth data in healthy infants.
• The method demonstrated good accuracy in capturing the spatiotemporal dynamic changes of the highly folded cortex.
• Inter-subject correspondences and diffeomorphic temporal evolution trajectories were estimated during training.

Conclusions:

• The novel framework offers an unprecedented solution for predicting infant cortical development from a single baseline scan.
• This approach advances longitudinal neuroimaging analysis by effectively utilizing spatiotemporal learning.
• The findings contribute to a better understanding of early brain structural dynamics in infants.