Tractography Processing with the Sparse Closest Point Transform.

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Medical Image Analysis

Background:

• Diffusion MRI tractography reconstructs white matter pathways but lacks standardized processing algorithms.
• Existing methods are often customized, limiting the integration of general machine learning (ML) techniques.

Purpose of the Study:

• To develop a novel vector-space representation for tractography data using the sparse closest point transform (SCPT).
• To enable the application of existing ML algorithms for tractography processing tasks like clustering, simplification, and selection.

Main Methods:

• The SCPT involves extracting sparse landmarks and transforming curves relative to these landmarks.
• Applied SCPT to fiber bundle clustering (non-parametric k-means), simplification (redundancy reduction), and population-based selection (one-class Gaussian classifier).
• Evaluated performance against alternative methods, random subsampling, and manual selection using metrics like reliability and sensitivity to aging.

Main Results:

• SCPT facilitates effective and efficient tractography processing by bridging the gap between tractography data and ML algorithms.
• Demonstrated successful application in clustering, simplification for visualization, and automated bundle selection across a population.
• Showcased scan-rescan reliability and sensitivity to normal aging in bundle selection tasks.

Conclusions:

• The SCPT provides a versatile framework for applying established ML tools to diffusion MRI tractography.
• This novel approach enhances the analysis and utility of white matter pathway reconstruction.
• Openly available data and software support further research and application in neuroimaging analysis.