Self-supervised learning with application for infant cerebellum segmentation and analysis

Area of Science:

• Neuroimaging
• Developmental Neuroscience
• Artificial Intelligence in Medicine

Background:

• Accurate infant cerebellum segmentation is vital for understanding early neurodevelopment.
• Existing methods face challenges due to cortical folding, low contrast, and data heterogeneity.

Purpose of the Study:

• To propose an accurate self-supervised learning framework for infant cerebellum segmentation.
• To characterize early cerebellar development and identify factors influencing growth.

Main Methods:

• Developed a self-supervised learning framework for infant cerebellum segmentation.
• Validated the framework's accuracy across 358 subjects from three independent datasets.

Main Results:

• The first six postnatal months show the most rapid and dynamic cerebellar changes.
• Gray matter (GM) volume significantly influences cerebellar growth more than white matter (WM).
• Males exhibit larger GM and WM volumes than females; autistic males show larger volumes than neurotypical males.

Conclusions:

• The proposed framework enables accurate infant cerebellum segmentation.
• Early cerebellar development is dynamic, with distinct patterns observed in males and autistic individuals.
• Wider application of this method will advance research into cerebellar structure and function in typical and atypical development.