Samira M Kazan1, Laurentius Huber2, Guillaume Flandin1, Peter Bandettini2, and Nikolaus Weiskopf1,3
1Wellcome Trust Centre for Neuroimaging, UCL Institute of Neurology, University College London, London, United Kingdom, 2Functional Imaging Methods Laboratory of Brain, National Institute of Mental Health, Washington, DC, United States, 3Department of Neurophysics, Max Planck Institute for Human Cognition and Brain Sciences, Leipzig, Germany
Synopsis
We recently presented a
vascular autocalibration method (VasA) to account for vascularization
differences between subjects and hence improve the sensitivity in group
studies. Here, we validate the novel calibration method by means of direct comparisons
of VasA with the established measure of vascular reactivity, the M-value,
obtained during induced hypercapnia. We show strong evidence that VasA is
dominated by local vascular reactivity variations similarly to the M-value. We
conclude that the VasA calibration method is an adequate tool for application
in group studies to help increasing the statistical significance and reflects
to a large degree local vascularization.Purpose:
Statistical power of fMRI group-studies is
significantly hampered by high inter-subject-variance, arising from differences
in baseline physiology (i.e. blood volume-CBV). We recently presented a Vascular-Autocalibration
method (VasA) [1] to account for vascularization differences between subjects,
thus improving the sensitivity in group-studies. VasA is based on the observation that global slow-respiration induced
BOLD-signal-changes within an fMRI-experiment can be taken as an indicator for
vascular reactivity and baseline venous CBV. VasA calibration values can be
obtained from any fMRI-time-series, by estimating the low-frequency components
of the residuals in the task-GLM. These residuals resemble those fMRI-signal
variations that do not match up with the task-paradigm but are dominated from
variations in breathing patterns. Here, we investigate the mechanism and
the physiological basis underlying VasA. We compare it to the Davis’ model calibration-parameter M [2]. M
is a function (among others) of the baseline-CBV and venous-deoxyhemoglobin
concentration of the blood [3]. To make the VasA method available widely, we developed
a SPM-toolbox that readily integrates VasA into standard fMRI-processing.
Methods:
To compare VasA and the quantitative calibration
parameter M, we conducted two experiments in five subjects after receiving
their consent. The first experiment consisted of a hypercapnia task of
breathing air/5%CO2/air for 2min/5min/5min, respectively to estimate M. The heart-rate and
respiratory gas composition were recorded during the gas challenge. In the
second experiment, a 10-min flashing checkerboard paradigm (30s-rest vs. 30s-stimulation)
was used to activate the visual-cortex. During both experiments, time-series of
CBV-changes and BOLD-signal-changes were captured with the SS-SI-VASO sequence
[4]. The acquisition parameters were: 7T-Siemens-MRI scanner, 7-slices, TE/TI1/TI2/TR=19/765/2265/3000ms,
adjusted inversion efficiency=75% with tr-FOCI-pulse. To minimize and assess
the influence of partial-voluming
with WM and CSF, 1.5mm was used. The M-value was estimated with the Davis’ model [2] on a voxel-wise-basis
assuming: CBV
rest=5.5%, α
total =0.38, α
veins=0.2,
β=1 at 7T [5]. To estimate the VasA maps, the subjects’ time-series was fitted
voxel-wise using the respective GLM describing the experimental paradigm in SPM
[6]. The maps were extracted from the residuals of the model fit as described
in [1]. For direct comparison of VasA and M-values, the MRI-volume data were
coregistered to each other after the application of 3mm smoothing in spm.
Results:
Fig.1B shows the
correlation of a VasA with an M-value map in one representative participant. Fig.1A shows
the corresponding scatter density plot across all voxels. The correlation
coefficient was 0.83±0.15 (mean±SD) including all GM-voxels. The stability of this high correlation
between regional VasA and M-values across participants can be seen in Fig.2. Example
maps in Fig.1B-C indicate that regional contrast-to-noise-ratio (CNR) in VasA maps
is higher compared to M-value maps. The vascular reactivity shown in VasA maps was
relatively homogeneously distributed across GM and not only confined to the
visual cortex, where most of task response was located (Fig.1D). Even though
VasA captured vascular reactivity throughout large portions of GM, there was a
small tendency in VasA to overestimate vascular reactivity in regions of
significant CSF partial-voluming (defined in EPI space, based on multi-TI-T1-maps,
Fig.1E). This resulted in a non-linear trend for high M-values in the scatter
plots (curved arrows-Fig.2) which mainly reflected voxels containing pial-veins
close to CSF. Fig.3 depicts screenshots of the novel SPM-toolbox for VasA
analysis, to facilitate the use of VasA in group-studies.
Discussion:
The strong correlation between VasA and M-values suggests
that both measures have similar physiological origins. The deviations in areas
of large-partial-voluming with WM might be coming from the difficulty to
estimate M-value with low-SNR CBV-data in WM. The deviations in areas of large-partial-voluming
with CSF might arise from VasA overestimations due to increased cardiac/respiratory
noise contributions in CSF. The higher CNR of VasA compared to conventional
M-value maps is likely due to a higher-CNR of the gradient-EPI to the BOLD
effect compared to the low-CNR CBF/CBV imaging methods. VasA activation maps generally
do not show increased bias in areas of large task-related activation. This
suggests that VasA measures truthfully reflect vascular-reactivity rather
than residuals due to imperfect modelling of task demands or residual non
modelled neuronal activity.
Conclusion:
The data show a strong correlation between vascular
calibration measures obtained with VasA and the more established vascular
reactivity value M. This suggests that VasA-calibration maps reflect vascular
reactivity, particularly baseline venous CBV distribution. Since potential VasA
contaminations from inaccurate task modelling could not be detected, VasA
calibration is a reliable tool for fMRI calibration with enhanced CNR
compared to conventional M-value calibration. A SPM-Toolbox (Fig.3) was
developed and will be made available widely to help in analysing existing large
datasets in an efficient fashion and provide significantly increased
statistical-power.
Acknowledgements
The Wellcome Trust Centre for Neuroimaging is
supported by core funding from the Wellcome Trust 091593/Z/10/Z. The research leading to these results has received funding from the European Research Council under the European Union's Seventh Framework Programme (FP7/2007-2013) / ERC grant agreement n° 616905. References
1. Kazan S. K. et al. NeuroImage, 2015; 124:794-805
2.
Davis T. L. et al. PNAS, 1998;
95:1834–1839
3. Hoge
R. D. et al. MRM, 1999; 42:849–863
4.
Huber L. et al. MRM, 2014; 72:137-148
5.
Blockley N. et al. NMR Biomed, 2013;
26:987-1003
6.
Friston et al. Human Brain Mapping, 1995; 2:189-210