Elsevier

NeuroImage

Volume 35, Issue 2, 1 April 2007, Pages 625-634
NeuroImage

Model-based analysis of rapid event-related functional near-infrared spectroscopy (NIRS) data: A parametric validation study

https://doi.org/10.1016/j.neuroimage.2006.11.028Get rights and content

Abstract

To validate the usefulness of a model-based analysis approach according to the general linear model (GLM) for functional near-infrared spectroscopy (fNIRS) data, a rapid event-related paradigm with an unpredictable stimulus sequence was applied to 15 healthy subjects. A parametric design was chosen wherein four differently graded contrasts of a flickering checkerboard were presented, allowing directed hypotheses about the rank order of the evoked hemodynamic response amplitudes. The results indicate the validity of amplitude estimation by three main findings (a) the GLM approach for fNIRS data is capable to identify human brain activation in the visual cortex with inter-stimulus intervals of 4–9 s (6.5 s average) whereas in non-visual areas no systematic activation was detectable; (b) the different contrast level intensities lead to the hypothesized rank order of the GLM amplitude parameters: visual cortex activation evoked by highest contrast > moderate contrast > lowest contrast > no stimulation; (c) analysis of null-events (no stimulation) did not produce any significant activation in the visual cortex or in other brain areas.

We conclude that a model-based GLM approach delivers valid fNIRS amplitude estimations and enables the analysis of rapid event-related fNIRS data series, which is highly relevant in particular for cognitive fNIRS studies.

Introduction

Event-related (ER) experimental designs are useful and often a precondition in examining cognitive processes since they are much more flexible than blocked task paradigms (Burock et al., 1998, Rosen et al., 1998, Pollmann et al., 2000). By them (a) a randomized stimulus order can be presented to the subject providing for a class of paradigms which is “naturally” event-related (e.g. oddball paradigm), (b) trials can be analyzed based on the co-registered behavioral data (e.g. ratings of stimuli, false/correct responses to stimuli) and (c) dynamics of the evoked activity can be analyzed.

Compared to EEG studies of event-related potentials (ERPs), techniques like functional magnetic resonance imaging (fMRI) or functional near-infrared spectroscopy (fNIRS) deal with hemodynamic signals which are inherently slow. The hemodynamic response (HR) evoked by a brief stimulus returns to baseline after 10–12 s or more (Boynton et al., 1996, Buckner et al., 1996) compared to the millisecond range of signals in EEG. As a consequence of these slow signals, stimuli that are separated less than one HR cycle result in a temporal overlap of the recorded HR. This, in turn, hampers a straightforward time-locked averaging procedure as it can be performed in the majority of evoked potential studies (overlap problems in ERP studies are discussed in Woldorff, 1993). Therefore, using sufficiently long inter-stimulus intervals (ISIs) is one solution to prevent overlapping HRs (Buckner et al., 1996) and thereby allows event-related averaging analogously to the analysis of non-overlapping ERPs. However, this approach is inefficient regarding its utilization of experimental time: less stimuli can be presented in a fixed time period which results in a loss of statistical power. Furthermore, uncontrollable confounding states (boredom) or activities (exploration behavior) of the subject might occur during these long delays (Serences, 2004).

In order to perform rapid ER designs which apply ISIs shorter than a complete HR cycle, adequate analysis procedures which handle temporal overlaps of the signals have to be applied. Provided that successive HRs summarize linearly (sufficient linearity of the HRs has been demonstrated with ISIs of 2 s in Dale and Buckner, 1997) and are time invariant, overlapping responses from rapid ER protocols can be analyzed in terms of the general linear model (GLM) by explicitly molding overlapping responses. Therefore, a reasonable hemodynamic response function (HRF) is convolved with the stimulation protocol (delta function) and taken as the predictor for the functional time series. This model-based analysis approach according to the GLM has become the standard strategy for analyzing functional data in the fMRI domain (Bullmore et al., 1996, Friston et al., 1995, Worsley and Friston, 1995).

Regarding the fNIRS domain, such a standard strategy for data analysis, as first proposed by Schroeter et al. (2004b), has not been fully established until today although there are several similarities to fMRI: the data basis (hemodynamic responses) and the possible experimental designs are highly comparable. Mirroring the development of fMRI research, the traditional fNIRS study designs of the past (block-design protocols) have been recently complemented by ER designs (e.g. Horovitz and Gore, 2004, Izzetoglu et al., 2005, Jasdzewski et al., 2003, Kennan et al., 2002, Plichta et al., 2006a, Plichta et al., 2006b, Schroeter et al., 2002). This development will result in a need of an adequate statistical analysis procedure for rapid ER-fNIRS designs. Of course the GLM approach is only one possibility beside many other reasonable analysis approaches, and the particular research question has to dictate the use of either GLM or other approaches. Anyhow, in the event that the GLM approach is adequate for data analysis, the fNIRS community can profit from the rich experiences of the fMRI domain and utilize the general framework of a model-based analysis approach. However, until today rapid ER-fNIRS studies which employ a GLM for data analysis are rare. Izzetoglu et al. (2005) performed a model-based ER analysis of fNIRS data where the ISIs were shorter than an HR cycle. In that study, participants had to solve anagrams while frontal brain activation was recorded by multi-channel fNIRS. The results indicate a parametric modulation of the fNIRS response due to the task difficulty. However, only oxyhemoglobin (O2Hb) data are described. No results are reported in terms of the regional specificity of the hemodynamic response and, thus, systemic effects cannot be excluded (Franceschini et al., 2003). Furthermore, the stimulus order was not randomized and no null control (i.e. no stimulation condition) was included. In another study of Boas et al. (2003), an event-related motor paradigm was conducted where the mean ISI was 8 s (range 2 s–33s) and a GLM approach was applied. However, a validation of the estimated beta weights, e.g. by conducting a parametric experimental design, was not the scope of the latter study.

The aim of the present work is to test if the GLM approach is capable of detecting significant human brain activation in a rapid ER-fNIRS design (replicating the findings of Izzetoglu et al., 2005) with a randomized stimulus order and, more crucially, if the GLM estimates of the amplitudes are valid regarding their relative height and correspond to varying stimulus intensities. A simple checkerboard paradigm was chosen because of its strong and robust activation effects in the primary visual cortex accessible with fNIRS (Plichta et al., 2006b). A parametric design with four levels of visual stimulation intensities is conducted. The visual stimuli vary regarding their luminance contrast levels (0%, 8%, 40% and 97%) and were chosen based on findings from fMRI studies (Avidan et al., 2002, Boynton et al., 1999, Heeger and Ress, 2002, Tootell et al., 1998) which have demonstrated that an increase of stimulus contrast levels results in a monotonic increase of the primary visual cortex (V1) activity.

Section snippets

Subjects and stimulation protocol

A total of 15 subjects were examined (mean age = 25.3 ± 2.6 years). We included both males and females (7 males; 8 females) regardless of hair color or handedness. All subjects had normal or corrected to normal vision. No subject had a history of any neurological disorder. All subjects were informed about the nature of the experiment as well as the operating mode of the NIRS instrument, before giving written informed consent. A brief instruction to remain relaxed and to avoid any major body

Behavioral data

All presented checkerboard stimuli were followed by button presses in each subject, indicating that the subjects remained attentive during the whole experiment. Subjective ratings by the participants indicate that the chosen contrast levels were clearly distinguishable intra- and inter-individually: the entire set of checkerboards was correctly classified by 11 subjects, one misclassification (error rate = 0.6%) occurred in one subject and two misclassifications (error rate = 1.2%) were made by two

Discussion

The aim of the present study was to examine if a model-based time series approach as performed in fMRI ER data analysis delivers estimates of fNIRS amplitudes which mirror a known rank order of differently graded stimulation intensities. Therefore, a rapid ER design with a parametric modulation of the stimulus intensity was conducted. The results show that the model-based analysis approach according to the GLM is capable of detecting event-related human brain activity recorded with fNIRS in the

Conclusion

The present study shows that the GLM framework of statistical analyses as applied in the fMRI domain can be expanded to the fNIRS domain. The GLM approach delivers valid fNIRS amplitude estimations and enables the analysis of rapid event-related fNIRS data series, which is highly relevant in particular for cognitive fNIRS studies. Moreover, the effective application of a GLM based analysis in the present study may facilitate a straightforward and intuitive understanding of fNIRS results for

Acknowledgments

The authors would like to thank: Hitachi Medical Corporation for the ETG-4000 equipment; Thomas Meigen for letting us use his MAVO-MONITOR®, Thomas Kammer for his helping comments regarding the visual stimulation; Antje B.M. Gerdes for the fruitful discussion of the experimental design; Andreas J. Bartsch, Vince D. Calhoun and the staff of the Bender Institute of Neuroimaging (B.I.O.N.), University of Giessen (director: Prof. Dr. Dieter Vaitl) for generously sharing their analysis expertise.

References (38)

  • M.L. Schroeter et al.

    Towards a standard analysis for functional near-infrared imaging

    NeuroImage

    (2004)
  • M.L. Schroeter et al.

    Investigating the post-stimulus undershoot of the BOLD signal—A simultaneous fMRI and fNIRS study

    NeuroImage

    (2006)
  • J.T. Serences

    A comparison of methods for characterizing the event-related BOLD timeseries in rapid fMRI

    NeuroImage

    (2004)
  • K.J. Worsley et al.

    Analysis of fMRI time-series revisited-again

    NeuroImage

    (1995)
  • G. Avidan et al.

    Contrast sensitivity in human visual areas and its relationship to object recognition

    J. Neurophysiol.

    (2002)
  • D.A. Boas et al.

    Can the cerebral metabolic rate of oxygen be estimated with near-infrared spectroscopy?

    Phys. Med. Biol.

    (2003)
  • G.M. Boynton et al.

    Linear systems analysis of functional magnetic resonance imaging in human V1

    J. Neurosci.

    (1996)
  • R.L. Buckner et al.

    Detection of cortical activation during averaged single trials of a cognitive task using functional magnetic resonance imaging

    Proc. Natl. Acad. Sci. U. S. A.

    (1996)
  • E. Bullmore et al.

    Statistical methods of estimation and inference for functional MR image analysis

    Magn. Reson. Med.

    (1996)
  • Cited by (229)

    • fNIRS: Non-stationary preprocessing methods

      2023, Biomedical Signal Processing and Control
    View all citing articles on Scopus
    View full text