Labeling lateral prefrontal sulci using spherical data augmentation and context-aware training.

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Brain Anatomy

Background:

• Cortical sulcal labeling traditionally focuses on deep sulci, neglecting shallow tertiary sulci due to annotation scarcity and high variability.
• Automated methods are needed to accurately map both deep and shallow sulcal regions for comprehensive brain analysis.

Purpose of the Study:

• To develop an automated framework for precise regional labeling of primary, secondary, and tertiary sulci in the lateral prefrontal cortex (LPFC).
• To address the challenge of neuroanatomical variability in sulcal patterns using novel data augmentation and training strategies.

Main Methods:

• Utilized spherical convolutional neural networks for sulcal labeling.
• Implemented a novel surface data augmentation technique synthesizing data through rigid to non-rigid deformation trajectories.
• Developed a two-stage, context-aware training process prioritizing primary/secondary sulci before tertiary sulci inference.

Main Results:

• The proposed data augmentation significantly improved labeling accuracy for both deep and shallow sulci across pediatric and adult datasets.
• Context-aware training further enhanced the accuracy of shallow (tertiary) sulci labeling compared to data augmentation alone.
• The method outperformed conventional multi-atlas and baseline spherical convolutional neural network approaches.

Conclusions:

• The developed automated framework effectively labels both deep and shallow cortical sulci, overcoming limitations of existing methods.
• The combination of surface data augmentation and context-aware training offers a robust solution for neuroanatomical variability in sulcal pattern analysis.
• This work provides valuable tools for understanding LPFC neuroanatomy and its functional organization, applicable to broader cortical regions.