An Example-Based Multi-Atlas Approach to Automatic Labeling of White Matter Tracts.

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Medical Image Analysis

Background:

• Accurate classification of white matter (WM) tracts into anatomic bundles is crucial for understanding brain connectivity.
• Existing methods often struggle with individual variability in WM tract shapes and trajectories.
• Manual classification is time-consuming and prone to inter-rater variability.

Purpose of the Study:

• To develop an automated, example-based, multi-atlas approach for classifying white matter tracts.
• To address the challenge of individual variability in brain anatomy.
• To improve the efficiency and accuracy of white matter tract classification.

Main Methods:

• An example-based multi-atlas strategy was employed, leveraging expert-provided data.
• Multiple atlases were constructed to capture inter-subject variability in bundle shapes.
• A voting scheme facilitated multi-atlas data exploitation, with WM tracts grouped by shape similarity for simultaneous labeling.
• The approach was implemented on a graphics processing unit (GPU) for enhanced computational efficiency.

Main Results:

• The approach demonstrated high classification performance via nested cross-validation.
• Average sensitivities for left and right hemisphere bundles were 89.5% and 91.0%, respectively.
• Average false discovery rates were 14.9% for the left and 14.2% for the right hemisphere.

Conclusions:

• The proposed multi-atlas method effectively automates white matter tract classification.
• The approach successfully accounts for individual variability in brain anatomy.
• This GPU-accelerated technique offers a significant improvement in efficiency and accuracy for neuroimaging analysis.