Fiberprint: A subject fingerprint based on sparse code pooling for white matter fiber analysis.

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Human Anatomy

Background:

• Diffusion magnetic resonance imaging (dMRI) is crucial for white matter characterization.
• Individual variations in white matter structure, function, and geometry necessitate personalized analysis methods.
• Existing methods may not fully capture the unique geometric properties of white matter fibers across individuals.

Purpose of the Study:

• To develop a subject fingerprint, termed Fiberprint, for quantifying individual uniqueness in white matter geometry.
• To leverage sparse coding of fiber trajectories for creating a compact, bundle-wise feature vector.
• To demonstrate the efficacy of Fiberprint in individual identification and twin/sibling studies.

Main Methods:

• Learning a sparse coding representation for white matter fiber trajectories.
• Mapping fiber trajectories to a common space defined by a dictionary.
• Generating a subject fingerprint using a pooling function on fiber trajectories within bundles.
• Analyzing data from 861 Human Connectome Project subjects.

Main Results:

• Fiberprint, based on approximately 3000 fiber trajectories, can uniquely identify individuals.
• The method successfully aids in twin/sibling identification, aligning with existing twin study findings.
• The generated feature vector (dimension of 50) effectively captures white matter fiber geometry variability.

Conclusions:

• Fiberprint provides a powerful and computationally efficient framework for characterizing individual white matter geometry.
• This approach enhances the analysis of large neuroimaging datasets by offering a compact representation of unique white matter features.
• Fiberprint has significant potential for applications in personalized medicine and understanding neurological variations.