Voxel-wise detection of functional networks in white matter

Abstract

Functional magnetic resonance imaging (fMRI) depicts neural activity in the brain indirectly by measuring blood oxygenation level dependent (BOLD) signals. The majority of fMRI studies have focused on detecting cortical activity in gray matter (GM), but whether functional BOLD signal changes also arise in white matter (WM), and whether neural activities trigger hemodynamic changes in WM similarly to GM, remain controversial, particularly in light of the much lower vascular density in WM. However, BOLD effects in WM are readily detected under hypercapnic challenges, and the number of reports supporting reliable detections of stimulus-induced activations in WM continues to grow. Rather than assume a particular hemodynamic response function, we used a voxel-by-voxel analysis of frequency spectra in WM to detect WM activations under visual stimulation, whose locations were validated with fiber tractography using diffusion tensor imaging (DTI). We demonstrate that specific WM regions are robustly activated in response to visual stimulation, and that regional distributions of WM activation are consistent with fiber pathways reconstructed using DTI. We further examined the variation in the concordance between WM activation and fiber density in groups of different sample sizes, and compared the signal profiles of BOLD time series between resting state and visual stimulation conditions in activated GM as well as activated and non-activated WM regions. Our findings confirm that BOLD signal variations in WM are modulated by neural activity and are detectable with conventional fMRI using appropriate methods, thus offering the potential of expanding functional connectivity measurements throughout the brain.

Keywords: BOLD; Functional activation; Visual stimulation; White matter; fMRI.

Fig. 1.

Distributions of the magnitude at the stimulus frequency in BOLD signals in WM and GM during visual stimulation. The MSF is thresholded at 0.4 MMSF for both WM and GM. Right side of the images is the subject’s left. Yellow arrows point to the OR, and the green arrow points to the V1.

Fig. 2.

Polarities of BOLD responses in WM and GM. Red and blue color denotes positive and negative correlation with the mean time series of activated voxels in the V1 respectively. Yellow contours demarcate WM boundaries.

Fig. 3.

Signal waveforms and frequency spectra in activated voxels of OR (red) and V1 (green) during visual stimulation. Voxels in the OR and V1 with MSF above 0.4 MMSF were defined as activated voxels. The BOLD time courses in the top row are average intensities of the twelve subjects studied, and the time courses in the middle row are average percent changes in signal intensities. Black curves in the top and bottom rows are the stimuli waveforms and frequency spectra. Stimulus-induced signal increases from resting period were 1.78% and 5.29% for the OR and V1 respectively.

Fig. 4.

Distributions of the magnitude corresponding to the frequency of the visual stimulation in a resting state. The threshold level is the same as in the stimulation condition. Right side of the images is the subject’s left. Yellow arrows point to supra-threshold clusters.

Fig. 5.

T-maps of WM and GM signals during visual stimulation. The t-statistic threshold is T>1.96 (p<0.05) for WM and T>5 (p<10⁻⁶) for GM. Cluster size threshold is 0 for both WM and GM. No corrections for multiple comparisons were performed.

Fig. 6.

Distributions of probability density of fibers in the visual system. Twelve WM bundles in each subject were tracked from their skeletons using probabilistic fiber tracking from DTI data. The probability density is the average of the twelve subjects studied.

Fig. 7.

Dice similarity coefficient between the MSF and PDF maps with different combinations of threshold levels.

Fig. 8.

Variations of Dice coefficient with the group size. The abscissa is the size of the subject group, and the ordinate is Dice coefficient between the average MSF map and the PDF maps. For each group size n, the average MSF map was calculated from all possible n-combinations from the set of twelve subjects studied. Dotted lines are the corresponding standard deviation, which were computed from the set of individual Dice coefficients between the MSF of each n-combination and the PDF maps.

Fig. 9.

Comparisons of BOLD signal profiles among three representative brain regions. (A) Mean signal intensity, (B) Standard deviation, and (C) Coefficient of variations in BOLD time series. From left to right in panels (A-C) are activated GM region, activated and non-activated WM regions. Blue and green bars denote resting state and visual stimulation respectively; error bars denote standard deviations of the measurements; asterisks (*) and (**) denote p<0.05 and p<0.01 from paired, two tailed t-tests respectively. Stripes in the green bars denote the portions measured from 3^rd–10^th TRs after the onset of stimulation. The number of voxels for computing statistics of the activated GM, WM and non-activated WM regions are respectively 3980 (all activated voxels), 1430 (all activated voxels) and 1460 (down sampled).