Training shortest-path tractography: Automatic learning of spatial priors.

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Medical Image Analysis

Background:

• Diffusion MRI tractography is standard for white matter tract delineation.
• Current methods often require manual post-processing and expert knowledge to remove false positives.
• This limits the feasibility of large-scale clinical tractography studies.

Purpose of the Study:

• To develop and validate a novel shortest-path tractography framework incorporating automatically learned prior spatial information.
• To demonstrate the framework's ability to improve tract delineation robustness and accuracy.
• To reduce reliance on manual expert intervention in tractography analysis.

Main Methods:

• Implemented a shortest-path tractography algorithm enhanced with learned prior spatial information.
• Developed methods for automatically generating priors from population data.
• Validated the approach using Human Connectome Project data and data from a clinical scanner.
• Compared results against a reference atlas and visual inspection.

Main Results:

• The learned prior significantly increased the overlap of tractography output with a reference atlas on both high-quality and clinical datasets.
• Visual inspection confirmed improved tract delineation robustness.
• Priors learned from high-quality data improved results on lower-quality clinical data.

Conclusions:

• Automatic incorporation of learned prior knowledge into tractography enhances accuracy and robustness.
• This framework reduces the need for expert interactive delineation, improving the feasibility of large clinical tractography studies.
• The method shows promise for more reliable and efficient white matter tract analysis in research and clinical settings.