Multi-output Gaussian processes for enhancing resolution of diffusion tensor fields.

Area of Science:

• Medical Imaging
• Computational Neuroscience
• Biophysics

Background:

• Diffusion tensor (DT) fields are crucial for clinical applications like brain connectivity and disease diagnosis.
• Low spatial resolution in diffusion magnetic resonance imaging (dMRI) data limits the accurate representation of small tissue fibers.
• Interpolation is a key technique for improving the spatial resolution of DT fields.

Purpose of the Study:

• To develop and evaluate a novel feature-based interpolation method for enhancing the spatial resolution of DT fields.
• To address the limitations of low-resolution dMRI data in capturing fine tissue structures.
• To improve the accuracy and preserve the properties of interpolated DT fields.

Main Methods:

• Extraction of diffusion tensor features: logarithm of eigenvalues (direction) and Euler angles (orientation).
• Application of multi-output Gaussian processes (MOGP) assuming a common covariance matrix for interpolated features.
• Reconstruction of new diffusion tensors from interpolated eigenvalues and Euler angles.

Main Results:

• The proposed MOGP method achieves higher accuracy compared to state-of-the-art DT interpolation techniques.
• Performance is comparable to methods using Generalized Wishart processes for positive semidefinite matrix interpolation.
• The MOGP approach successfully preserves essential diffusion tensor properties, such as fractional anisotropy.

Conclusions:

• Feature-based interpolation using MOGP is an effective method for enhancing the spatial resolution of DT fields in dMRI.
• This technique offers improved accuracy and preserves critical biophysical information, advancing clinical applications.
• MOGP provides a robust alternative for high-fidelity diffusion tensor field reconstruction.