White Matter Tract Segmentation as Multiple Linear Assignment Problems.

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Medical Image Analysis

Background:

• Diffusion MRI (dMRI) generates tractograms representing brain white matter streamlines.
• Tract segmentation organizes streamlines into meaningful anatomical tracts for neurosurgical planning and tractometry.
• Supervised methods leverage prior knowledge for more reliable tract segmentation.

Purpose of the Study:

• To present a novel supervised tract segmentation method using multiple examples as prior information.
• To improve streamline correspondence by formulating it as a linear assignment problem (LAP).
• To enhance the accuracy and anatomical meaningfulness of segmented brain tracts.

Main Methods:

• Developed a supervised tract segmentation method based on streamline correspondence.
• Formulated streamline correspondence as a linear assignment problem (LAP), enforcing one-to-one mapping.
• Combined the Jonker-Volgenant algorithm (LAPJV) with efficient nearest neighbor computation.
• Introduced a ranking strategy to merge correspondences from multiple examples.

Main Results:

• The LAP-based segmentation method demonstrated vastly superior accuracy compared to ROI-based methods.
• The proposed method showed substantial accuracy improvements over the nearest neighbor (NN) strategy.
• Validation on Human Connectome Project (HCP) datasets confirmed the method's effectiveness.

Conclusions:

• The LAP-based approach offers a more accurate and robust solution for supervised tract segmentation.
• This method effectively addresses limitations of previous streamline correspondence techniques.
• A Free/Open-Source implementation is provided, facilitating wider adoption and research.