Inter-network correlations between white-matter networks measured using BOLD fMRI

1786 

Abstract Submission 

Nayana Menon¹, Jonathan Polimeni², J. Jean Chen³

¹University of Toronto, Oakville, Ontario, ²Athinoula A. Martinos Center for Biomedical Imaging, Massachusetts General Hospital, Charlestown, MA, ³Baycrest Health Sciences, Toronto, Ontario

Nayana Menon  
University of Toronto
Oakville, Ontario

Jonathan Polimeni  
Athinoula A. Martinos Center for Biomedical Imaging, Massachusetts General Hospital
Charlestown, MA

J. Jean Chen  
Baycrest Health Sciences
Toronto, Ontario

Detecting brain activation white matter (WM) using blood-oxygenation level-dependent (BOLD) fMRI data is a topic of growing interest (1-2). Recent work using BOLD has demonstrated that BOLD activity in WM corresponds to task demands (3). Additionally, the detection of functional networks in the WM was demonstrated using BOLD fMRI (REF). It has also been found that signals in WM networks correlate with signals in functional gray matter (GM) networks, which have established connections (4). Furthermore, inter-network connections have been established amongst GM functional networks (4). While these findings fill gaps in the knowledge around the interpretation of WM fMRI signals and networks, they also highlight that little is known about whether inter-network connections exist in the WM. This work aims to determine whether WM networks display spatial clusters of inter-network correlations in the same way that GM networks do.

Participants
Here we report findings from the 10 youngest healthy participants (equal males and females) selected from the Human Connectome Project's Aging dataset (mean age = 36.6 years) (5).
fMRI Data Acquisition
10 youngest participants from the HCP Aging study (HCP-A) were chosen. fMRI scans were acquired with simultaneous multi-slice gradient-echo EPI (TR/TE = 800/37 ms, flip angle = 52°, 184 frames) and 2.0 mm isotropic voxels that covered the whole brain. During the fMRI scans, participants were to memorize names for a series of faces, beginning with an 22 s encoding block: where the faces and names were shown followed by a 2 second cue to memorize followed by 5 face/name pairs shown for 4s each (6). This is followed by a distractor block of 2 seconds followed by a 20 second GO/NoGO task. This is followed by the recall block with a 2s cue and 20s where the faces are shown again. This is repeated twice for each subject (6).
Data Processing
The raw BOLD data were preprocessed using FSL for motion correction, slice-timing correction, then normalized and demeaned such that all values in the time series are provided in units of %BOLD. The data were also registered to MNI space. The data were then masked into WM and GM, and the WM data for each subject was further denoised using independent component analysis (ICA) with 30 components being generated and noisy components being visually determined and removed (7). Masks were generated for the WM and GM networks of the brain using the functionnectome atlas (8), and the GM regions that overlapped with WM were removed from the WM masks, then the masks were eroded by 1 voxel to fully separate the masked regions.
Analysis
The mean time series of the GM and WM regions were extracted using the masks derived from the functionnectome atlas. The time series within the WM were correlated with each other for each subject using Pearson's correlation coefficient, then the r-value matrices were converted using the Fisher transform to z-standardized matrices and averaged to generate a group-level matrix.

The WM functional connectivity matrix of averaged z-scores is shown in Fig. 1, thresholded at z>2. We can observe that there are high correlations between the activity in distinct WM networks identified from the functionnectome atlas, that are similar to the inter-network connectivity patterns among the corresponding cortical GM regions As shown in Fig. 2, these networks are in deep WM and do not overlap with GM networks.

This study presents an initial investigation of connections between WM networks identified from the functionnectome atlas in fMRI task data. We were able to identify WM networks that were significantly correlated with each other during visual and motor task conditions, and that, by nature of the functionnectome atlas, were separate from GM networks, with a far smaller likelihood of GM effects on the WM analysis.

Activation (eg. BOLD task-fMRI)

Connectivity (eg. functional, effective, structural)

fMRI Connectivity and Network Modeling ¹

White Matter Anatomy, Fiber Pathways and Connectivity ²

BOLD fMRI

Data analysis

FUNCTIONAL MRI

Statistical Methods

White Matter

WHITE MATTER IMAGING - DTI, HARDI, DSI, ETC