Superficial white matter analysis: An efficient point-cloud-based deep learning framework with supervised contrastive...

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Machine Learning

Background:

• Diffusion MRI tractography maps in vivo brain white matter connections.
• White matter parcellation is crucial for quantifying and visualizing tractography data.
• Existing methods struggle with complex superficial white matter (SWM) parcellation.

Purpose of the Study:

• To introduce Superficial White Matter Analysis (SupWMA), a novel deep learning framework.
• To achieve efficient and consistent parcellation of 198 superficial white matter clusters.
• To address limitations of current methods in SWM analysis.

Main Methods:

• A two-stage deep learning framework utilizing a point-cloud-based network.
• Supervised contrastive learning for improved streamline representation and outlier detection.
• Training on a large-scale dataset with labeled SWM clusters and implausible streamlines.

Main Results:

• SupWMA achieved highly consistent and accurate SWM parcellation across six independent datasets.
• Demonstrated excellent generalization across different ages and health conditions, including neonates and tumor patients.
• Significantly faster computational speed compared to state-of-the-art methods.

Conclusions:

• SupWMA provides a robust and efficient solution for superficial white matter parcellation.
• The framework's performance highlights its potential for broad applications in neuroscience research and clinical settings.
• SupWMA advances the analysis of brain connectivity, particularly in complex SWM regions.