The fluctuations in physiological signals (cardiac rhythm and respiration) are included in blood–oxygen-level-dependent (BOLD) functional MRI (fMRI) data. As the brain is working, the physiology, which includes cerebral blood flow, cerebral blood volume, and cerebral metabolism, plays with the mechanisms of BOLD. Unlike thermal and machine noise of Gaussian distribution, non-Gaussian physiological signals induce considerable fluctuations in BOLD signals because they are caused by the same mechanism 1. In the spatial analysis of resting state fMRI, physiological signals contribute more in gray matter and venous sinuses than in white matter regions 2. When temporal correlation analysis of BOLD time series was carried out for functional connectivity, physiological signals mixed with BOLD signals were derived from neuronal activity. Because of the limited Nyquist frequency using a low temporal resolution of fMRI with a repetition time of 1–2s, the physiological signals aliased to unsettled frequency bands below fMRI resolution 3, and the aliased physiological noises could not be removed by a band-pass filter.
Many methods have been used to attenuate the effect of physiological signals in low temporal resolution fMRI 1,3–10. Some studies have ignored the physiological effects by independent component analysis (ICA) using regressors derived from tissues without gray matter 9,11,12. However, the physiological signals transmit to the entire brain, not just in the cerebrospinal fluid and white matter. With respiratory and cardiac noise sources contributing about 10% toward functional connectivity in fMRI 13, the interference of physiological signals in functional or effective connectivity should be validated.
In the study, the approach of time-lag effective connectivity was used to validate the effect of physiological signals by detecting known action networks. The functional network connectivity (FNC)-based connectivity will be compared before and after the correction of physiological interference by retrospective image-based correction (RETROICOR). The goal of the study is to verify the physiological interference in the noninvasive detection of an effective network using BOLD-based fMRI with a suboptimal temporal resolution.
Materials and methods
Thirty-six normal right-handed individuals without a history of Diagnostic and Statistical Manual of Mental Disorders, 4th ed. axis I disorders, major physical illness, neurological diseases, substance abuse, head trauma, or metal implants in the body were recruited. They filled in the modified Edinburgh handedness inventory for laterality index and signed the informed consent form (institutional review board of the Taipei Veterans General Hospital) before the fMRI experiments and practiced the action task before scanning.
The participants were asked to perform the motor task by visual cuing through a mirror mounted on the head coil during the fMRI study. Two kinds of figures were provided as congruent (>>>>>) and incongruent (>><>>) with an exposure time of 5000 ms. The participants were asked to press the buttons of the MR-compatible plate using the middle or index fingers for congruent or incongruent trials, respectively, as soon as possible. The congruent and incongruent trials were pseudorandomized at a ratio of 88 : 8, and interstimulus intervals were 6400±200 ms, with jitters from 200 to 600 ms during fixation periods on cross. The stimuli were presented by Presentation version 0.71 (NeuroBehavioral Systems, San Francisco, California, USA).
Echo planar imaging (EPI) and T1-weighted structural images were acquired by a 3 T scanner (MR 750; General Electric, Waukesha, Wisconsin, USA) equipped with an eight-channel head coil. The parameters of T2*-weighted EPI were repetition time (TR)/echo time (TE)=2000/30 ms, voxel size=3.6×3.6×4 mm3, repetition number=405, and the parameters of T1-weighted structural image were TR/TE/inversion time=6.2/2.3/450 ms, voxel size=0.9×0.9×0.9 mm3. ECG (vectorcadiogram), pulse gating, and respiration were recorded by MR-compatible devices of an MR scanner at digitization frequencies of 1000, 100, and 25 Hz, respectively.
The processes of data analyses included (a) calculation of the action task for the criteria of participant selection, (b) image preprocessing I and II for selected participants, (c) RETROICOR for removing physiological signals, (d) group ICA for before or after RETROICOR, and (e) FNC analysis of the action network.
The RT, SD, and error rate of all congruent trials in the action task were calculated. Twenty-five participants with an error rate less than 10% were included for the FNC analysis.
The image preprocessing using SPM8 (Statistical Parametric Mapping 8; Functional Imaging Laboratory, Wellcome Department of Imaging Neuroscience, UCL, London, UK) included (I) slice timing, realignment, and de-scalp (brain extraction tool; FMRIB Software Library, FMRIB Centre, Oxford, UK) and (II) coregistration, normalization, and smooth with full-width at half-maximum of an 8×8×8 mm Gaussian kernel. Correlates of the action task were modeled by an event-type paradigm using the timing of response convolved with a hemodynamic function at the first level, and a one-sample t-test at the second level [family-wise error (FWE)-corrected, P<0.005].
PG was used to represent signals of cardiac rhythm in this study because ECG data were easily influenced by electromagnetic field and dynamic gradients with variations in the R–R interval and signal amplitude. After image preprocessing I, RETROICOR was performed using the in-house programming by Matlab (version 7.6; MathWorks Inc., Natick, Massachusetts, USA). The Fourier spectrum of full-time series was calculated for each voxel, and changes in spectral power were obtained by summing the magnitude spectra at the peaks of the spectral component (±0.03 Hz) after RETROICOR processing for both PG and respiration 5. The unit of power spectrum was decibel (dB).
fMRI analysis and component selection ICA
After image preprocessing and RETROICOR, 25 datasets were available for two groups with and without RETROICOR processing, respectively. All 50 datasets were processed with group ICA by GIFT (group ICA of fMRI tool box, v1.3h; The Mind Research Network, Albuquerque, New Mexico, USA). Activation maps of each component were calculated using a one-sample t-test and converted into a z score. From 59 components estimated by the minimum description length criteria, six components were selected by the known action network model 14–16, and included visual cortices (component A), parietal lobe (component B), supplementary motor area (SMA)/premotor cortex (component C), left and right motor/sensory cortices (components D and E), and ventricles (component F). The correlates of the selected components were defined using the time courses of selected components as regressors for every participant at the first level and group analyses using a one-sample t-test at the second level in statistical analyses of SPM8.
FNC of selected components by ICA
The connectivity and direction between two components among the six components were calculated and interpolated the time course to 50 ms bins 12 by FNC. The lag time between two components was detected by shifted time courses from −3 to +3 s for obtaining the maximal correlation coefficient 11,12. The direction of connectivity was decided by the shifted time of the maximal correlation coefficient. Each pair-wised connection (e.g. A–B) was tested using a one-sample t-test within each group without or with RETROICOR. The differences between two groups were tested using a paired t-test (the maximal correlation coefficient of the group without RETROICOR subtracted from the other group with correction). To verify the known effect of physiological signals, the component E (right motor cortex) was replaced by a component F (ventricles) when degrees of freedom were fixed.
Twenty-five participants, 24.4±3.0 years old and male/female=12/13, were selected from the original database of 36 participants according to (a) the error rate of congruent trials and (b) gender balance. The mean and SD of RT of the action task was 0.55±0.10 and 0.14±0.10 s, respectively. The mean of the laterality index by the modified Edinburgh handedness questionnaire was 79.8±14.9 for right-handed individuals. There was no significant difference between sex and age/RT/SD/laterality index (P>0.05).
Effects of RETROICOR
On comparing the spectral power before and after RETROICOR, there was a significant difference, with P-values being 0.0001 and 0.0003 for PG and respiration signals, respectively, by paired t-tests. The ranges of the main frequency were 0.02–0.25 Hz, with individual variations among participants. The mean and SD of reduced spectral powers of PG and respiration were 0.95±0.82 and 0.3±0.38, respectively.
Functional map and FNC maps of the action networks
Group fMRI results of the congruent trial showed the activation of the left precentral gyrus (BA 4), left somatosensory cortices (BA 1), bilateral visual cortices, right cerebellum, left parietal lobe, left putamen, and right medial globus pallidus (Fig. 1). Left lateralization of BOLD-based activation was found (one-sample t-test, FWE-corrected, P<0.005).
On the basis of group results of selected components derived from group ICA, spatial correlates were verified for components A [spatial correlates and voxels (left/right cerebral hemisphere): BA18 (1611/1038), BA19 (1105/2765), and BA1 (7393/408)], B [BA40 (747/619) and BA7 (444/307)], C [BA6 (1248/741) and BA3 (388/29)], and E [BA2 (39/272), BA3 (55/162) BA4 (11/80), and BA40 (48/186)] by including all 25 participants (FWE-corrected, P<0.0001; voxel extension >100). Correlates of components D [BA2 (441/40), BA3 (176/75), BA4 (185/24), and BA40(141/2)] and F [lateral ventricle (70/126)] were identified by 15 and 16 participants, respectively, for group analysis because of individual variations (P<00.005; voxel extension >10).
The FNC of the action network that included five components (A, B, C, D, and E) with and without RETROICOR is shown in Fig. 2a and b. The arrow and color indicate the direction of effective connectivity and lag time between two components, respectively. Functional connectivity strength of the action task was the minimal P-values of the connectivity in a one-sample t-test among all the selected components (yellow numbers in the Fig. 2). Connections of B→D, A→D, and D→E (→: represent the effective connectivity between two modules) were detected after RETROICOR (P<0.05, one-sample t-test), and E→C was changed to C→E. The physiological effects on A→D, B→C, and B→D are shown in Fig. 2c by comparing FNC with and without RETROICOR (paired t-test, P<0.05). By replacing component E by component F, the connectivity of F→A, D→F, and C→F related to component F (ventricles) disappeared after RETROICOR (Fig. 2d and e).
Compatible spatial components and effective connectivity of the action network were detected by group ICA and time-lag FNC by selecting suitable ICA components. On the basis of previous knowledge of an external cuing network in action, the effective connectivity of visuomotor task consisted of visual cortex→parietal lobe→SMA/premotor cortex→contralateral primary motor cortex 14,16,17. Bilateral superior and inferior parietal lobules (SPL and IPL) of component B were consistent with the fMRI study of unilateral finger tapping 15.
Effects of physiological interference were shown by (a) a significant difference in the spectral power of physiological signals without and with RETROICOR, (b) group difference among the visual cortex (component A), SPL/IPL (component B), SMA/premotor cortex (component C), and left primary motor/sensory cortices (component D) (Fig. 2c by a paired t-test, P<0.05), and (c) more explicable results of FNC with RETROICOR.
Three connectivities, including left→right primary motor/sensory cortices (D→E), visual cortex→left primary motor/sensory cortices (A→D), and SPL/IPL→left primary motor/sensory cortex (B→D), were found after RETROICOR. The connection between bilateral primary motor/sensory cortices (D→E) was consistent with interhemispheric inhibition as shown by the interhemispheric inhibition using repetitive transcranial magnetic stimulation and fMRI 18,19. Inverted connectivity from E→C to C→E after RETROICOR echoes interhemispheric inhibition (Fig. 2b). The connections of A→D/A→E and B→D/D→E, which were not reported in previous studies, implied alternative paths of the action performance as a network. The present results of time-lag FNC detected the effective connectivity or networks that were consistent with previous studies 16,17. The effect of physiological signals misled the interpretation for the connectivity without preprocessing of RETROICOR.
After RETROICOR, the falsified connectivity among the ventricle and the other components was evidently attenuated. With significant physiological effects in our ICA-based approaches without and with RETROICOR (Fig. 2d and e), the insufficiency of ICA in differentiating physiological signals from task-related BOLD-based responses was clear 11,12.
The limitations of the present study were as follows: (a) temporal resolution of the present fMRI study with TR of 2 s may limit the spatial decomposition of spatial ICA, for example merging SMA with premotor areas as a single component C and (b) selection of ICA components can be modified by a spatial template derived from the functional map or fixed ICA component numbers 11.
The time-lag FNC of a dedicated event-related fMRI of action has validated time-lag causality analysis of a known action network and the interference of physiological signals in the analyses of effective connectivity. In terms of the influence of physiological signals and the effect of RETROICOR on time-lag FNC, it can be concluded that there is (a) a physiological interference in the analyses of effective connectivity and (b) ICA is not useful for differentiating physiological signals from task-related BOLD-based responses.
The authors acknowledge the grant support of the National Science Council of Taiwan (NSC 99-2314-B-075-027) and the Taipei Veterans General Hospital (V100E3-005). The authors also acknowledge Dr Jen-Chuen Hsieh for his inspiring discussion.
Conflicts of interest
There are no conflicts of interest.
1. Chang C, Glover GH. Relationship between respiration, end-tidal CO2
, and BOLD signals in resting-state fMRI. Neuroimage. 2009;47:1381–1393
2. Kruger G, Glover GH. Physiological noise in oxygenation-sensitive magnetic resonance imaging. Magn Reson Med. 2001;46:631–637
3. Kiviniemi V, Ruohonen J, Tervonen O. Separation of physiological very low frequency fluctuation from aliasing by switched sampling interval fMRI scans. Magn Reson Imaging. 2005;23:41–46
4. Biswal B, DeYoe AE, Hyde JS. Reduction of physiological fluctuations in fMRI using digital filters. Magn Reson Imaging. 1996;35:107–113
5. Glover GH, Li TQ, Ress D. Image-based method for retrospective correction of physiological motion effects in fMRI: RETROICOR. Magn Reson Med. 2000;44:162–167
6. Lund TE, Madsen KH, Sidaros K, Luo WL, Nichols TE. Non-white noise in fMRI: does modelling have an impact? Neuroimage. 2006;29:54–66
7. Kelley DJ, Oakes TR, Greischar LL, Chung MK, Ollinger JM, Alexander AL, et al. Automatic physiological waveform processing for FMRI noise correction and analysis. PloS one. 2008;3:e1751
8. Lin FH, Nummenmaa A, Witzel T, Polimeni JR, Zeffiro TA, Wang FN, et al. Physiological noise reduction using volumetric functional magnetic resonance inverse imaging. Hum Brain Mapp. 2011 [Epub ahead of print] doi: 10.1002/hbm.21403
9. McKeown MJ, Hansen LK, Sejnowsk TJ. Independent component analysis
of functional MRI: what is signal and what is noise? Curr Opin Neurobiol. 2003;13:620–629
10. Sarkka S, Solin A, Nummenmaa A, Vehtari A, Auranen T, Vanni S, et al. Dynamic retrospective filtering of physiological noise in BOLD fMRI: DRIFTER. Neuroimage. 2012;60:1517–1527
11. Jafri MJ, Pearlson GD, Stevens M, Calhoun VD. A method for functional network connectivity
among spatially independent resting-state components in schizophrenia. Neuroimage. 2008;39:1666–1681
12. Assaf M, Jagannathan K, Calhoun V, Kraut M, Hart J Jr., Pearlson G. Temporal sequence of hemispheric network activation during semantic processing: a functional network connectivity
analysis. Brain Cogn. 2009;70:238–246
13. Cordes D, Haughton VM, Arfanakis K, Carew JD, Turski PA, Moritz CH, et al. Frequencies contributing to functional connectivity in the cerebral cortex in ‘resting-state’ data. Am J Neuroradiol. 2001;22:1326–1333
14. Gazzaniga MS, Ivry RB, Mangun GRGazzaniga M, Ivry R, Mangun G. The control of action. Cognitive neuroscience: the biology of the mind. 2009 New York W. W. Norton & Company Inc.:257–311
15. Horenstein C, Lowe MJ, Koenig KA, Phillips MD. Comparison of unilateral and bilateral complex finger tapping-related activation in premotor and primary motor cortex. Hum Brain Mapp. 2009;30:1397–1412
16. Kayser AS, Sun FT, D'Esposito M. A comparison of granger causality and coherency in fMRI-based analysis of the motor system. Hum Brain Mapp. 2009;30:3475–3494
17. Lin FH, Agnew JA, Belliveau JW, Zeffiro TA. Functional and effective connectivity
of visuomotor control systems demonstrated using generalized partial least squares and structural equation modeling. Hum Brain Mapp. 2009;30:2232–2251
18. Kobayashi M, Theoret H, Pascual-Leone A. Suppression of ipsilateral motor cortex facilitates motor skill learning. Eur J Neurosci. 2009;29:833–836
19. Hamzei F, Dettmers C, Rzanny R, Liepert J, Buchel C, Weiller C. Reduction of excitability (inhibition) in the ipsilateral primary motor cortex is mirrored by fMRI signal decreases. Neuroimage. 2002;17:490–496