Purpose

Our aim was to explore the consistently reported patterns of apparent positive and negative activations, derived via independent-component-analysis (ICA), in high-resolution functional QSM (fQSM^1-3) acquired at 7-Tesla and to compare them with fMRI results, using phase-contrast-angiograms for identification of large vessels. Statistical maps were generated from magnitude, phase, and susceptibility data collected during visual and motor task, and resting-state experiments.

Materials & Methods

Experiment Gradient-echo-2D-EPI (TR=3s, TE=25ms, FA=850, SENSE=3.5, voxel-dimensions=1.25x1.25x1.3mm, reconstructed-matrix=176x176x34) images of 4 consenting volunteers were acquired at 7-Tesla (Philips). The scan was repeated to acquire whole-brains with 160 slices. T1-Contrast-Anatomical-Scan 3D-inversion-recovery gradient-echo, TR=8.2ms, TE=3.79ms, FA=80, voxel-dimensions=0.94x0.94x1mm, Phase-Contrast-Angiography (PCA) 3D-Gradient-echo, TR=35.9ms, TE=7ms, FA=100, voxel-dimensions=0.9x0.9x0.9mm, VENC=15cm/sec. The paradigm4 required volunteers to first fix a projected cross for 9s. Subsequently, 10 blocks of upper-left/lower-right (ULLR) and upper-right/lower-left (URLL) color-changing wedges were presented over 200 EPI-scans. For motor-task, they were asked to perform a simple button-response task, i.e. right button press once they see the cross pointing right, or vice versa (event-related). Resting-state experiments were acquired with the same sequence parameters as the task experiment. QSM-Processing Phase data were unwrapped via the Laplacian approach⁵, and background-field-corrected via a variant of SHARP with variable spherical kernels (V-SHARP^{6, 7}, Rmax=12.5mm, threshold=0.012mm^-1). By means of field to susceptibility dipolar-inversion, QSM images were generated using an LSQR⁸, and a threshold-based k-space division (TKD)⁹ algorithms. A temporal-band-pass-filter [0.01-0.11Hz] was applied to fQSM data. Functional-Processing The sign of QSM data was inverted and the minimum value over all-time series was subtracted from each pixel value for QSM and SHARP data. The processed series of phase, QSM and magnitude data were normalized to MNI-space with 1.8mm isotropic-voxel-size, smoothed with a 2.5mm-Gaussian-kernel, and denoted as fPHASE, fQSM and fMRI, respectively. The data were subject to the spatial-domain ICA (FAST-ICA, GIFT¹⁰), allowing 20 components. Number of voxels in thresholded z-score maps were counted, in volumes-of-interest, i.e. in the visual cortex, in the thresholded-PCA, and in automatically segmented gray-matter.

Results

Single-voxel time-courses of the raw-phase showed large discontinuities that were eliminated during unwrapping and background-field-removal. While the SHARP-signal time-courses faintly resembled that of the magnitude, there was a much better similarity with that of the (-)QSM data (Fig. 1). Figure-2 shows patterns of positive and negative activations in fQSM (ULLR, |Z|>1.2). Activation maps for fPHASE (first row, hot & blue), fQSM (second and third row, hot & blue), and fMRI (green) are compared in Figure-3. Positive (hot) and negative (blue) patterns were also observed in z-score maps derived from corrected phase data (top, Fig. 3). After dipolar-inversion (fQSM), the area of apparent activation was more focused in the region expected according to the paradigm (white dotted-circle, Fig. 3). Some activations in fPHASE data that did not exist in fMRI data were eliminated by the dipolar inversion (white arrows, fPHASE). The total activated area (hot + blue) in fQSM data matched better with the area of positive fMRI activation (green), than it did in fPHASE. A slightly more confined activated area was observed in LSQR derived fQSM results vs. TKD. Figure-3 shows single-voxel signal time-courses for voxels (thick arrows), i.e. one voxel with positive z-scores in both fQSM (hot) and fMRI and one voxel with a negative z-score in fQSM (blue) and positive in fMRI. Components related to the visual and motor network are shown for fMRI (green, Z>1.2), fQSM (positive (hot), negative (blue), |Z|>1.2), and overlaid with thresholded-PCA (Figure 4b-c). For this specific subject, the ratio of commonly activated voxels of fQSM (|Z|>1.2) and fMRI (Z>1.2) with respect to the number of voxels in fQSM was ~40%, for both ULLR and URLL stimulation. The percentage of activated voxels in thresholded-PCA regions vs. total activated voxels was ~30%, and in the segmented gray-matter regions ~65%, for both fQSM (|Z|>1.2) and fMRI (Z>1.2). As observed in task cases, for resting state experiment, the area of visual network activation in fMRI (Z>1.2) is separated into regions of positive (Z>1.2) and negative (Z<-1.2) activation in fQSM (Fig.5).

Discussion

ICA-based fQSM analysis detected areas of significant positive and negative activations within areas of positive fMRI activation both in task-based and resting-state experiments. In contrast, there were no voxels with such a statistically significant negative activation in the fMRI z-score maps. Understanding the underlying reason for the differentiation of the apparent homogeneous positive fMRI activation in positive and negative fQSM activations needs further investigation, potentially in combination with other methods, such as measuring blood flow to understand counterbalancing effects to oxygenation. Furthermore, if it is of physiological origin, fQSM might offer unique opportunities, e.g., for disentangling variations in the hemodynamic response and neuronal activation.

Acknowledgements

No acknowledgement found.

References

1. Balla DZ, Sanchez-Panchuelo RM, Wharton SJ, Hagberg GE, Scheffler K, Francis ST et al. Functional quantitative susceptibility mapping (fQSM). NeuroImage 2014; 100: 112-24. 2. Ozbay PS, Warnock G, Rossi C, Kuhn F, Akin B, Pruessmann KP et al. Probing neuronal activation by functional quantitative susceptibility mapping under a visual paradigm: A group level comparison with BOLD fMRI and PET. NeuroImage 2016; 137: 52-60. 3. Chen Z, Liu J, Calhoun VD. Susceptibility-based functional brain mapping by 3D deconvolution of an MR-phase activation map. Journal of neuroscience methods 2013; 216(1): 33-42. 4. Kasper L, Haeberlin M, Dietrich BE, Gross S, Barmet C, Wilm BJ et al. Matched-filter acquisition for BOLD fMRI. NeuroImage 2014; 100: 145-60. 5. Schofield MA, Zhu Y. Fast phase unwrapping algorithm for interferometric applications. Optics letters 2003; 28(14): 1194-6. 6. Schweser F, Sommer K, Deistung A, Reichenbach JR. Quantitative susceptibility mapping for investigating subtle susceptibility variations in the human brain. NeuroImage 2012; 62(3): 2083-100. 7. Özbay PS, Deistung A, Feng X, Nanz D, Reichenbach JR, Schweser F. A comprehensive numerical analysis of background phase correction with V-SHARP. NMR in biomedicine 2016. 8. Li W, Wu B, Liu C. Quantitative susceptibility mapping of human brain reflects spatial variation in tissue composition. NeuroImage 2011; 55(4): 1645-56. 9. Özbay PS, Rossi C, Kocian R, Redle M, Boss A, Pruessmann KP et al. Effect of respiratory hyperoxic challenge on magnetic susceptibility in human brain assessed by quantitative susceptibility mapping (QSM). NMR in biomedicine 2015; 28(12): 1688-96. 10. Calhoun VC. Group ICA of fMRI Toolbox (GIFT). Available at: http://icatb.sourceforge.net. 2004.