Diffusion-informed spatial smoothing of fMRI data in white matter using spectral graph filters

Area of Science:

• Neuroimaging
• White Matter Analysis
• Functional Magnetic Resonance Imaging (fMRI)

Background:

• Functional magnetic resonance imaging (fMRI) studies have historically focused on gray matter (GM), largely overlooking white matter (WM) due to uncertainties about the blood oxygenation level-dependent (BOLD) contrast in WM.
• Emerging evidence confirms the functional relevance of the WM BOLD signal, highlighting its anisotropic spatio-temporal correlations and structure-specific fluctuations that mirror cortical activity.
• Conventional isotropic Gaussian filters are insufficient for denoising WM fMRI data due to the anisotropic nature of the BOLD signal, which is closely tied to axonal structure and orientation.

Purpose of the Study:

• To develop and validate an anisotropic spatial filtering scheme tailored for smoothing fMRI data specifically within white matter.
• To address the limitations of isotropic filters in capturing the anisotropic BOLD signal characteristics in WM.
• To enhance the sensitivity and specificity of white matter activation mapping in fMRI studies.

Main Methods:

• A graph-based representation of white matter (WM) was created using diffusion-weighted MRI data to encode local axonal structure and anisotropy.
• Subject-specific spatial filters were designed based on graph signal processing principles, adapting to individual WM architecture at each location.
• The proposed anisotropic filters were applied to smooth fMRI data in WM, serving as an alternative to standard isotropic Gaussian smoothing.

Main Results:

• Simulated phantom studies demonstrated that the proposed anisotropic filtering method exhibits superior sensitivity and specificity for detecting slender anisotropic activations compared to isotropic Gaussian filters.
• Application to the Human Connectome Project dataset revealed the method's capability to detect streamline-like activations within axonal bundles across seven functional tasks.
• The anisotropic filtering approach effectively denoises fMRI data in WM while preserving functionally relevant anisotropic signal characteristics.

Conclusions:

• The developed anisotropic spatial filtering scheme provides a more accurate and sensitive approach for analyzing functional brain activity in white matter using fMRI.
• This method overcomes the limitations of conventional smoothing techniques, enabling better detection of WM functional signals.
• The findings support the functional significance of the WM BOLD signal and offer a valuable tool for advancing white matter neuroimaging research.