An experimental evaluation of diffusion tensor image segmentation using graph-cuts.

Area of Science:

• Neuroimaging
• Medical Image Analysis
• Computational Neuroscience

Background:

• Diffusion Tensor Imaging (DTI) data segmentation is complex due to overlapping distributions of DTI measures like fractional anisotropy (FA).
• Accurate segmentation is crucial for characterizing white matter (WM) microstructure and for WM tractography masking.
• Existing methods face challenges in reliably distinguishing between white matter (WM), gray matter (GM), and cerebrospinal fluid (CSF).

Purpose of the Study:

• To develop and evaluate a graph-cuts segmentation method for extracting WM, GM, and CSF from brain DTI data.
• To assess the accuracy of the proposed method against manual segmentation by an expert.
• To demonstrate the utility of DTI segmentation for microstructural analysis and tractography.

Main Methods:

• A two-phase graph-cuts segmentation approach was employed.
• Cerebrospinal fluid (CSF) was segmented using third eigenvalue (lambda(3)) maps.
• White matter (WM) regions were extracted from fractional anisotropy (FA) maps.

Main Results:

• The graph-cuts method achieved high segmentation accuracy on ten in vivo human DTI datasets.
• Average volume overlap (VO) accuracies were approximately 0.90 for WM, 0.77 for GM, and 0.88 for CSF.
• The method demonstrated robust performance compared to manual segmentation by an expert.

Conclusions:

• The graph-cuts segmentation method provides an accurate and reliable approach for segmenting brain tissues in DTI data.
• This technique facilitates improved characterization of white matter microstructural properties.
• The method is valuable for applications requiring precise segmentation, such as WM tractography.