Extrapolation of sparse tensor fields: application to the modeling of brain variability

Area of Science:

• Neuroscience
• Computational Anatomy
• Medical Image Analysis

Background:

• Modeling anatomical variability in the human brain is crucial for understanding neurodevelopment and disease.
• Existing methods face challenges in handling complex tensor data representing structural variations.

Purpose of the Study:

• To develop an original method for computing average sulcal curves and modeling brain structure variability.
• To introduce an affine-invariant Riemannian framework for tensor field computations.
• To analyze brain variability asymmetry and its potential implications for functional neuroimaging.

Main Methods:

• Utilized a dataset of 72 sulcal lines from 98 healthy human subjects.
• Modeled the second-order moment of sulcal distribution using a sparse field of covariance tensors.
• Applied an affine-invariant Riemannian framework to generalize radial basis function interpolation and harmonic diffusion partial differential equations to tensor fields.

Main Results:

• Generated a dense 3D variability map consistent with previous findings.
• Demonstrated the model's ability to recover missing information using leave-one-out tests.
• Identified significant asymmetries in anatomical variance, notably in primary language areas.

Conclusions:

• The proposed Riemannian framework effectively models brain structure variability and tensor fields.
• The findings suggest that anatomical variance asymmetry may influence the sensitivity of functional magnetic resonance imaging (fMRI) studies.
• This approach offers new tools for analyzing brain asymmetry and its functional consequences.