Stefan M Spann1, Doris Grössinger2, Christoph Stefan Aigner1,3, Josef Pfeuffer4, Guilherme Wood2,5, and Rudolf Stollberger1,5
1Institute of Medical Engineering, Graz University of Technology, Graz, Austria, 2Institute of Psychology, University of Graz, Graz, Austria, 3Physikalisch-Technische Bundesanstalt (PTB), Braunschweig and Berlin, Germany, 4Application Development, Siemens Healthcare, Erlangen, Germany, 5BioTechMed-Graz, Graz, Austria
Synopsis
Real-time neurofeedback (RT-NF) fMRI allows
the subjects to regulate their own brain activity by providing them a neurofeedback.
Functional ASL is perfectly suited for RT-NF studies due to the absolute quantification
of activation related changes in the cerebral blood flow (CBF). In this study we
implemented a real-time solution for ASL data processing and feedback
generation which includes the following steps: data acquisition, image reconstruction,
post-processing and neurofeedback presentation. The results of this RT-NF fASL
study show that subjects were able to learn to regulate their own brain activation
during a finger tapping experiment.
Introduction
Real-time (RT) BOLD-fMRI neurofeedback
(NF) is a well-established method to regulate the brain activity of subjects by
presenting them a feedback1,2. The neurofeedback allows the subjects learning
to influence their own brain function which was successfully applied in patients
with schizophrenia and auditory hallucinations3 or patients with
chronic pain4. Functional Arterial Spin Labeling (fASL) is a very
promising approach for studying the neural activation5 due to its
sensitivity to blood flow alterations5 and it has some important
advantages compared to BOLD fMRI which makes fASL perfectly suited for NF
studies: increased spatial accuracy6, higher intra-individual reproducibility6,
and a direct activation related absolute cerebral blood flow (CBF) change5.
So far only one study investigated the real-time processing of fASL data7.
However, that study focused only on the online monitoring of the ASL signal
without providing feedback to the subject7. In this study we
implemented a real-time solution for ASL data processing and feedback
generation. This pipeline includes the acquisition of data, image reconstruction,
post-processing and neurofeedback presentation during a finger tapping task to
guide the subject’s performance. The computation of all processing steps is
done within the repetition time and allows guiding the subject’s activation
intermediately. The RT-NF finger tapping experiment was performed on
5 subjects. The results of this study show that the subjects were able to learn
to control their own brain activation and the mean CBF in the activation area
increased from 73.7±18.8 ml/100g/min without feedback to 86.2±18.9 ml/100g/min with
feedback presentation.Methods
5 subjects were measured on a 3T MR system
(Skyra, Siemens Healthcare, Germany) after giving written informed consent. A prototype
pCASL sequence with 3D-GRASE readout was used with the following imaging parameters:
FOV = 192x192x96 mm³, 3x3x6 mm3 resolution, 16 slices, phase/slice
oversampling = 10/15%, slice-partial fourier (PF) = 6/8, phase-PF = 6/8,
EPI-factor = 51, TF = 14, TR/TE = 4000/23ms, LD/PLD = 1800/1700ms. A baseline
(BL) finger tapping experiment was conducted using a block-wise paradigm with 6
interleaved blocks (32 sec rest/task). The experiment consists of one BL run and
two feedback runs (FB1 and FB2). The BL run was used to locate the activation
area in the motor cortex and to create the mask for each subject.
Figure 1 shows an overview of the proposed real-time neurofeedback processing pipeline. After acquisition and reconstruction
on the scanner site the ASL DICOM image is transferred to workstation 1 (WS1).
On WS1 the ASL preprocessing steps including motion-correction, surround
subtraction8, spatial filtering, perfusion quantification9 and CBF-map visualization
are performed in MATLAB (MathWorks, Natick, MA, USA) using SPM1210
and in-house MATLAB scripts. The whole procedure (from image reconstruction to
presenting feedback) lasts 2s which is within the range of one TR (4s). After
each acquisition the calculated CBF-maps were updated on WS1 (Figure 2) to allow
monitoring the image-quality and CBF-values during the experiment. A GLM was
fitted after BL scan in the CBF-time series for localization of the activation
area in the motor cortex. The created mask serves as basis for FB1 and FB2. During
the feedback runs the change in CBF was presented to the subject via a monitor
to guide the cognitive activation. The subjects tried to modify the original
finger tapping experiment which was a continuous alternating finger tapping, by
changing the frequency, using different finger wiggles etc. to increase the
neural activity in the motor cortex. An increase in neural activity increases
the CBF and hence the feedback-bar (Figure 2B). The feedback representation was
implemented in PsychoPy software.Results and Discussion
Table 1 lists the
mean CBF-values in the defined activation region. Except for subject 3 we observe
a slight increase in CBF of FB1 compared to BL. It should be noted that different
strategies such as changing the frequency, using different finger wiggles etc. can
lead to higher as well as lower CBF values in the activation area. In
contrast, subject 3 was not able to find the right strategy in FB1. However, in
the second feedback run (FB2) the mean CBF clearly increased for all subjects
in the activation area. This indicates that the subjects were able to learn
strategies to regulate their own neural activity. The CBF increase is also visible
in the CBF-maps of each subject shown in Figure 3 (red arrows). Figure 4 shows
the mean CBF time course of all subjects averaged over the 6 blocks. The same increase
in CBF as for the individual subjects is observable with the highest increase during
FB2.Conclusion
We proposed a real-time
solution for fASL RT-NF studies where all necessary acquisition and processing
steps were executed within a single TR. This allows to monitor the ASL signal
and to guide the subjects’ cognitive process. The results of this study demonstrate
that subjects can learn to regulate their brain activity during a finger
tapping experiment based on the provided neurofeedback. This can be used to
promote rehabilitation of different symptoms e.g. in stroke patients were motor
and cognitive processes are often impaired. However only a few subjects were included in this proof of principle study and an evaluation on more subjects will be part of future work.Acknowledgements
NVIDIA Corporation Hardware grant support.References
1. Sitaram R, Ros T, Stoeckel L,
et al. Closed-loop brain training: the science of NF. Nature Reviews
Neuroscience 2017:18(2):86.
2. Weiskopf N, Veit R, Erb M, et al. Physiological self-regulation of
regional brain activity using real-time functional magnetic resonance imaging
(fMRI): methodology and exemplary data. Neuroimage. 2003:19(3):577-586.
3. Orlov ND, Giampietro V,
O'Daly O, et al. Real-time fMRI neurofeedback to down-regulate superior
temporal gyrus activity in patients with schizophrenia and auditory
hallucinations: a proof-of-concept study. Transl Psychiatry. 2018;8(1):46.
4. deCharms RC, Maeda F, Glover
GH, et al. Control over brain activation and pain learned by using real-time
functional MRI. Proc Natl Acad Sci U S A. 2005;102(51):18626–18631.
5. Detre JA, Leigh JS, Williams DS, et al.
Perfusion imaging. Magn Reson Med. 1992;23(1):37-45.
6. Raoult H, Petr J, Bannier E, et al.
Arterial spin labeling for motor activation mapping at 3T with a 32-channel
coil: reproducibility and spatial accuracy in comparison with BOLD fMRI.
Neuroimage. 2011;58(1):157-67.
7. Hernandez-Garcia L,
Jahanian H, Greenwald MK, et al. Real-time functional
MRI using pseudo-continuous arterial spin labeling. Magn Reson Med 2011;65:1570–1577.
8. Lu H, Donahue MJ and van Zijl PC.
Detrimental effects of BOLD signal in arterial spin labeling fMRI at high field
strength. Magn Reson Med 2006;56(3):546-52.
9. Buxton RB, Frank LR, Wong EC, et al. A
general kinetic model for quantitative perfusion imaging with arterial spin
labeling. Magn Reson Med 1998;40(3):383–396.
10. Friston KJ, Ashburner
J, Kiebel S, et al. Statistical parametric mapping: the analysis of funtional
brain images. Elsevier/Academic Press 2007.