Using a deep generation network reveals neuroanatomical specificity in hemispheres.

Area of Science:

• Neuroscience
• Computational Biology
• Medical Imaging

Background:

• Brain asymmetry is a fundamental organizational principle, yet its underlying mechanisms remain incompletely understood.
• Identifying neuroanatomical components unique to each hemisphere is crucial for deciphering the development of brain asymmetry.

Purpose of the Study:

• To develop and validate a computational model for quantifying brain asymmetry.
• To leverage deep generative networks (DGNs) for predicting one cerebral hemisphere from its counterpart.

Main Methods:

• Training DGNs to predict the contralateral hemisphere from a given hemisphere, thereby learning interhemispheric dependencies.
• Utilizing the difference between actual and predicted hemispheres to derive a metric for hemisphere-specific components.
• Assessing the biological plausibility and reliability of the DGN model and the proposed asymmetry metric.

Main Results:

• The DGN model successfully reconstructed homologous components, demonstrating biological plausibility.
• The developed metric for hemispheric specialization proved reliable and captured significant individual variations.
• The approach provides a novel tool for investigating brain asymmetry.

Conclusions:

• This work presents a promising self-supervised deep generative network approach for analyzing brain asymmetry.
• The proposed metric offers valuable insights into hemisphere-specific components influenced by genetic and environmental factors.
• The study contributes novel tools and understanding to the field of neuroanatomy and brain organization.