Elsevier

NeuroImage

Volume 45, Issue 4, 1 May 2009, Pages 1117-1125
NeuroImage

Cluster analysis of resting-state fMRI time series

https://doi.org/10.1016/j.neuroimage.2008.12.015Get rights and content

Abstract

Functional MRI (fMRI) has become one of the leading methods for brain mapping in neuroscience. Recent advances in fMRI analysis were used to define the default state of brain activity, functional connectivity and basal activity. Basal activity measured with fMRI raised tremendous interest among neuroscientists since synchronized brain activity pattern could be retrieved while the subject rests (resting state fMRI). During recent years, a few signal processing schemes have been suggested to analyze the resting state blood oxygenation level dependent (BOLD) signal from simple correlations to spectral decomposition. In most of these analysis schemes, the question asked was which brain areas “behave” in the time domain similarly to a pre-specified ROI.

In this work we applied short time frequency analysis and clustering to study the spatial signal characteristics of resting state fMRI time series. Such analysis revealed that clusters of similar BOLD fluctuations are found in the cortex but also in the white matter and sub-cortical gray matter regions (thalamus). We found high similarities between the BOLD clusters and the neuro-anatomical appearance of brain regions.

Additional analysis of the BOLD time series revealed a strong correlation between head movements and clustering quality. Experiments performed with T1-weighted time series also provided similar quality of clustering. These observations led us to the conclusion that non-functional contributions to the BOLD time series can also account for symmetric appearance of signal fluctuations. These contributions may include head motions, the underling microvasculature anatomy and cellular morphology.

Introduction

Functional MRI (fMRI) revolutionized the field of human brain mapping. Based on the fine equilibrium between blood oxygenation levels, areas of task-specific brain activity could be indirectly defined (Belliveau et al., 1991, Jezzard et al., 2002, Ogawa et al., 1990). Although not proven, the recent hypothesis on the biochemical and cellular mechanisms underlying the blood oxygenation changes following brain activity points to a cascade of events that occur following pre-synaptic activity (Logothetis et al., 2001, Raichle and Mintun, 2006). The fMRI signal was found to correlate with synchronous neuronal activity (Logothetis et al., 2001, Raichle and Mintun, 2006). Because most fMRI studies rely on comparisons between different tasks and resting periods, few questions were raised in the early days of fMRI about the contribution of the basal activity of the brain; for example, what happens to the fMRI signal when we assume that the brain rests? Based on PET studies, Raichle and colleagues (Raichle et al., 2001, Raichle and Mintun, 2006) defined a default network of brain activity consisting of areas that are metabolically active in the resting state. In BOLD fMRI, these areas demonstrate consistent attenuated BOLD signal concurrent with increased response at other locations (Gusnard et al., 2001, Gusnard and Raichle, 2001, Raichle et al., 2001, Raichle and Mintun, 2006, Vincent et al., 2007).

Resting state functional MRI was recently used to access basal brain activity (Beckmann et al., 2005, Biswal et al., 1995, Cordes et al., 2001, Damoiseaux et al., 2006, De Luca et al., 2006, Fox et al., 2005, Fransson, 2005, Greicius et al., 2003, Lowe et al., 1998, Peltier and Noll, 2002, Raichle and Mintun, 2006). In this type of experiment, the BOLD imaging conventionally used for task-related fMRI is acquired continuously while the subject is asked not to perform any specific task. With resting state fMRI data, it became evident that the brain can be segmented into regions of correlated BOLD signal fluctuations (Beckmann et al., 2005, Biswal et al., 1995, Damoiseaux et al., 2006, De Luca et al., 2006, Fox et al., 2005, Fransson, 2005, Greicius et al., 2003, Lowe et al., 1998, Nir et al., 2006, Peltier and Noll, 2002, Raichle and Mintun, 2006). This signal pattern was suggested to reflect a brain network of neurons working in concert (functional connectivity) (Beckmann et al., 2005, Cordes et al., 2001, Greicius et al., 2003, Lowe et al., 1998, Nir et al., 2006, Peltier and Noll, 2002), or spontaneous neuronal activity (Biswal et al., 1995, Damoiseaux et al., 2006, De Luca et al., 2006, Fox et al., 2005, Fransson, 2005). The latter was suggested to be related to the default brain network originally defined by PET studies (Fransson, 2005).

Conventional fMRI analysis extracts task-related BOLD signal changes by using a statistical comparisons between the time series and the task timings (Belliveau et al., 1991, Jezzard et al., 2002). In contrast, resting state fMRI analyzes the BOLD signal with no information about the timing of neuronal events, but rather uses only the similarity of fluctuations in different voxels. Thus, it can not separate blood oxygenation contributions (T2⁎ effect) from any other signal sources. Noise contribution to the BOLD signal has been thoroughly studied (Bodurka et al., 2007, Kruger and Glover, 2001). It has been demonstrated that a typical BOLD signal embraces several noise sources; non-physiological noises that result from the MRI scanner and physiological noises originated from the subjects. The physiological noises are divided into two main sources: functional and non-functional. The physiological noises were also shown to be weighted by the measured absolute signal (Kruger and Glover, 2001) implying that its fluctuations might be tissue specific.

A number of signal processing schemes have been suggested for analyzing the resting state BOLD signal, ranging from simple correlations to spectral decomposition of frequency analysis (Beckmann et al., 2005, Biswal et al., 1995, Damoiseaux et al., 2006, De Luca et al., 2006, Fox et al., 2005, Fransson, 2005, Greicius et al., 2003, Lowe et al., 1998, Peltier and Noll, 2002). Interestingly, frequency analysis revealed low-frequency fluctuations of the data (below 0.1 Hz) that are correlated between functionally related regions of the brain.(Biswal et al., 1995, Raichle and Mintun, 2006). In this work we wish to characterize these low frequency BOLD signals by "k-means" clustering analysis of the fluctuated signal and its spectral decomposition. Such analysis was performed on the cortex, sub-cortical regions (e.g. thalamus) and white matter. As input we used the original time series rest-fMRI signal as well as a short-time Fourier transform (spectrogram) that provides information not only on the frequencies of the fluctuating signal, but also on their amplitude changes over time. The functional relevance of the rest fMRI signal fluctuations was investigated by comparing the obtained clusters with anatomical and cyto-architectonic atlases (Talairach (Talairach and P., 1988) and Brodmann (Garey, 1999)). To define to what extent the rest fMRI signal fluctuations reflect tissue functionality, we repeated such analysis on non-BOLD MRI time series measurement and on BOLD white matter signal. We also investigated the effect of non-functional sources of physiological noise (i.e. head motion) to the observed rest fMRI clusters.

Section snippets

Data acquisition

Eleven healthy subjects aged 22–42 years underwent MRI in a 3 T scanner (GE, Milwaukee, USA) using an 8-channel head coil. The institutional review board (IRB) committee approved the research protocol and each subject signed an informed consent. The MRI protocol included high resolution anatomical images: T2-weighted images using a fast spin-echo sequence (TR/TE = 7000/85 ms, echo trail length of 16), fluid-attenuated inversion recovery (FLAIR) images (TR/TE/TI = 9000/102/2100 ms) and 3D spoiled

Results

In this work we used a spectrogram (short-time Fourier transform) analysis in order to characterize the time dependency of BOLD time series frequency variations. This analysis procedure was compared with conventional analysis on the time series. Fig. 1 shows typical spectrogram of one cortical voxel of one representative subject. We found that the amplitude of certain frequencies of the resting-state fMRI changes significantly with time (a 2-3 fold increase or decrease was frequently observed)

Discussion

The first, most prominent finding of this work is that the clustering analysis of multi-dimensional acquisition of BOLD and non-BOLD MRI signals allows segmentation of the brain into its neuroanatomical regions in both gray matter and white matter regions.

In this work we used clustering analysis to classify brain regions according to their fluctuated BOLD signal. Since we used the Euclidean distance as the factor of the cluster analysis, this procedure is similar, in principle, to the

Conclusions

Based on the anatomical appearance of the resting-state clusters and the abovementioned validation steps, we argue that the tissue microstructure and non-functionally related noise contributors influence correlated BOLD signal fluctuations. Without imposing functional paradigms on the BOLD time series, separating the functional contribution to the signal fluctuation from the anatomical/noise contributions is challenging. In view of the above results, the resting-state fMRI and functional

Acknowledgments

The authors wish to thank the following organizations for financial support: The Center for Complexity Sciences (AM), The Israel Science Foundation (YA) and The Levie-Edersheim-Gitter Institute for Functional Brain Mapping (OP). The authors also wish to thank that Raymond and Beverly Sackler institute for biophysics and the Israel science foundation (ISF) for the purchase of the 7T MRI scanner as well as the Strauss Center for Computational Neuroimaging of Tel Aviv University.

References (34)

  • TriantafyllouC. et al.

    Comparison of physiological noise at 1.5 T, 3 T and 7 T and optimization of fMRI acquisition parameters

    NeuroImage

    (2005)
  • BeckmannC.F. et al.

    Investigations into resting-state connectivity using independent component analysis

    Philos. Trans. R. Soc. Lond., B Biol. Sci.

    (2005)
  • BelliveauJ.W. et al.

    Functional mapping of the human visual cortex by magnetic resonance imaging

    Science

    (1991)
  • BiswalB. et al.

    Functional connectivity in the motor cortex of resting human brain using echo-planar MRI

    Magn. Reson. Med.

    (1995)
  • CordesD. et al.

    Frequencies contributing to functional connectivity in the cerebral cortex in “resting-state” data

    AJNR. Am. J. Neuroradiol.

    (2001)
  • DamoiseauxJ.S. et al.

    Consistent resting-state networks across healthy subjects

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

    (2006)
  • FoxM.D. et al.

    The human brain is intrinsically organized into dynamic, anticorrelated functional networks

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

    (2005)
  • Cited by (0)

    View full text