Diffusion maps clustering for magnetic resonance q-ball imaging segmentation.

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Medical Image Analysis

Background:

• White matter fiber clustering is crucial for neuroimaging research, aiding in atlas generation, visualization, and statistical analysis.
• Diffusion Tensor Imaging (DTI) faces limitations in complex white matter regions with crossing fibers.
• High-Angular Resolution Diffusion Imaging (HARDI), specifically Q-Ball Imaging (QBI), overcomes DTI limitations by reconstructing the diffusion orientation distribution function (ODF).

Purpose of the Study:

• To present a diffusion maps clustering method for segmenting complex white matter fiber bundles using QBI data.
• To demonstrate the advantages of diffusion maps clustering over traditional methods like N-Cuts and Laplacian eigenmaps.
• To validate the method's performance on both synthetic and real-brain datasets, particularly in regions with fiber crossings.

Main Methods:

• Utilized spherical harmonic representation of the ODF as input for diffusion maps clustering.
• Compared diffusion maps clustering with N-Cuts and Laplacian eigenmaps.
• Employed an adaptive scale-space parameter for automatic cluster number determination.

Main Results:

• Diffusion maps clustering requires fewer input data hypotheses and reduces segmentation artifacts compared to classical methods.
• The proposed method successfully segments fiber bundles and crossing regions in complex HARDI data, unlike DT-based methods.
• Results on a real-brain dataset demonstrate accurate segmentation of known white matter fiber bundles.

Conclusions:

• Diffusion maps clustering with ODF representation offers a robust and accurate method for white matter fiber segmentation, especially in complex neural pathways.
• The technique surpasses traditional diffusion tensor-based approaches in handling fiber crossings and reducing artifacts.
• This method advances neuroimaging analysis by enabling more precise segmentation for atlasing, visualization, and statistical studies.