Elsevier

NeuroImage

Volume 157, 15 August 2017, Pages 531-544
NeuroImage

Comparison of the spatial resolution of source imaging techniques in high-density EEG and MEG

https://doi.org/10.1016/j.neuroimage.2017.06.022Get rights and content

Highlights

  • Evaluation of intrinsic spatial resolution of EEG and MEG source imaging methods.

  • Comparison of a novel entropy-based technique (cMEM) to well-established methods.

  • cMEM showed similar localization error, lowest spatial spread and reduced crosstalk.

  • MEG and high-density EEG (256 sensors) presented a similar level of spatial accuracy.

  • Theoretical results confirmed with real data of somatosensory evoked potentials.

Abstract

Background

The present study aims at evaluating and comparing electrical and magnetic distributed source imaging methods applied to high-density Electroencephalography (hdEEG) and Magnetoencephalography (MEG) data. We used resolution matrices to characterize spatial resolution properties of Minimum Norm Estimate (MNE), dynamic Statistical Parametric Mapping (dSPM), standardized Low-Resolution Electromagnetic Tomography (sLORETA) and coherent Maximum Entropy on the Mean (cMEM, an entropy-based technique). The resolution matrix provides information of the Point Spread Functions (PSF) and of the Crosstalk functions (CT), this latter being also called source leakage, as it reflects the influence of a source on its neighbors.

Methods

The spatial resolution of the inverse operators was first evaluated theoretically and then with real data acquired using electrical median nerve stimulation on five healthy participants. We evaluated the Dipole Localization Error (DLE) and the Spatial Dispersion (SD) of each PSF and CT map.

Results

cMEM showed the smallest spatial spread (SD) for both PSF and CT maps, whereas localization errors (DLE) were similar for all methods. Whereas cMEM SD values were lower in MEG compared to hdEEG, the other methods slightly favored hdEEG over MEG. In real data, cMEM provided similar localization error and significantly less spatial spread than other methods for both MEG and hdEEG. Whereas both MEG and hdEEG provided very accurate localizations, all the source imaging methods actually performed better in MEG compared to hdEEG according to all evaluation metrics, probably due to the higher signal-to-noise ratio of the data in MEG.

Conclusion

Our overall results show that all investigated methods provide similar localization errors, suggesting very accurate localization for both MEG and hdEEG when similar number of sensors are considered for both modalities. Intrinsic properties of source imaging methods as well as their behavior for well-controlled tasks, suggest an overall better performance of cMEM in regards to spatial resolution and spatial leakage for both hdEEG and MEG. This indicates that cMEM would be a good candidate for studying source localization of focal and extended generators as well as functional connectivity studies.

Introduction

High-density Electroencephalography (hdEEG), defined here as EEG with 256 electrodes, and Magnetoencephalography (MEG) are two complementary and non-invasive neurophysiology modalities used to depict electromagnetic brain activity (Stefan, 2009, Ebersole and Ebersole, 2010, Ahlfors et al., 2011). Electrical and magnetic source imaging consist in solving a so-called inverse problem, localizing the generators of scalp EEG or MEG signals into the brain (Darvas et al., 2004).

The inverse problem is ill-posed by nature and a unique solution can only be found if specific constraints are added for regularization (Panday et al., 2009). Multiple inverse solution approaches are nowadays available, including equivalent current dipole fitting, dipole scanning approaches and distributed source models (see Wendel et al., 2009, Klamer et al., 2014).

The spatial accuracy of source imaging techniques is influenced by several factors, including the choice of modality (Pittau et al., 2014), the number of sensors (Sohrabpour et al., 2014), the orientation of the generator (Ahlfors et al., 2010), the conductivity of the biological tissues (Aydin et al., 2014) and the source imaging technique used to solve the inverse problem (Hauk et al., 2011).

The objective of this study was to compare the intrinsic spatial resolutions of different source imaging techniques applied to hdEEG and MEG signals, when considering these two modalities with approximately the same number of sensors. We analyzed the resolution matrix, whose application in the neuroimaging field was originally proposed by Menendez et al. (1997), to characterize the spatial properties of a linear inverse operator. The spatial resolution matrix is a square matrix, whose size is the number of dipolar sources, providing the following two features (Schoffelen and Gross, 2009):

  • 1.

    The columns of the resolution matrix quantify the Point Spread Functions (PSF) of every dipolar source of the source space. Each PSF is assessing the solution of the source imaging method for the activation a single cortical dipole, when considering noise-free data. Analyzing the localization error and the spatial extent of the PSF provides information about the intrinsic spatial property of a source imaging technique.

  • 2.

    The rows of the resolution matrix represent the crosstalk functions (CT) which reflect the influence a single dipolar source may have on the estimation of the generators in its neighborhood. Hence, the spatial extent of the CT informs on the amount of “source leakage” and the potential bias in the estimation of functional connectivity patterns leading to spurious local coherence (Schoffelen and Gross, 2009).

An ideal resolution matrix should be the identity matrix. In practice, any source estimate is subject to blurring (if large amplitude values are found in off-diagonal terms in the matrix) and mislocation (if off-diagonal values were higher amplitude than diagonal values).

Once PSF and CT maps are constructed for each dipolar source, one can assess their spatial properties through evaluation metrics such as Dipole Localization Error (DLE) and Spatial Dispersion (SD) (Liu et al., 2002, Molins et al., 2008). DLE measures the Euclidean distance between the maximum of the PSF or CT maps and the true source location, whereas SD quantifies the spatial spread around the true source location.

Beyond the theoretical analysis of the resolution matrix, a validation of the comparison between the intrinsic spatial resolution can be achieved by studying real data acquired under well-controlled paradigms, for which the location of generator is known a priori. To that motive, we measures somatosensory evoked responses measured after electrical stimulation of the median nerve. This paradigm is known to generate evoked response exhibiting the activation of the contralateral primary sensory hand region (Balzamo et al., 2004). In this context, for which the generator is located in a predefined focal brain region, properties of source imaging techniques and comparison between them could also be evaluated using DLE and SD metrics, as proposed by Molins et al. (2008).

We chose to evaluate and compare four distributed sources localization methods. Three of them are well-known linear operators: Minimum Norm Estimate (MNE) (Hämäläinen and Ilmoniemi, 1994), dynamic Statistical Parametric Mapping (dSPM) (Dale and Sereno, 1993) and standardized Low-Resolution Electromagnetic Tomography (sLORETA) (Pascual-Marqui, 2002). The fourth one is the coherent Maximum Entropy on the Mean (cMEM) (Amblard and Lapalme, 2004, Grova et al., 2006), which is a novel non-linear method specifically evaluated for its sensitivity to recover the spatial extent of the underlying cortical generators (Chowdhury et al., 2013, Heers et al., 2016, Grova et al., 2016). Whereas the calculation of the resolution matrix is straightforward for the linear methods (MNE, sLORETA and dSPM), specific estimation of the resolution matrix for the non-linear method cMEM was performed with the iterative reconstruction of the PSF of every dipolar source.

We proposed a systematic and quantitative assessment of these source imaging techniques based on two strategies: (i) theoretical analysis of the resolution matrix; (ii) study of the source estimated from hdEEG and MEG responses evoked by electrical median nerve stimulation, in the primary somatosensory cortex.

Section snippets

Subjects selection

Five right-handed healthy subjects (3 males, mean age ± standard deviation =26.6±3.21) were selected for this study. The study was approved by the Research Ethics Board of the Montreal Neurological Institute and Hospital and a written informed consent was signed by all participants prior to the procedures.

Electrical Median Nerve Stimulation

Electrical Median Nerve Stimulation (MNS) was performed using a Digitimer system (Digitimer DS7A, Letchworth Garden City, U.K). 600 stimuli were delivered to the left and right median nerves

Analysis of DLE and SD distributions

For each source localization technique and each modality, the distributions of DLE and SD metrics over all dipolar sources along the cortical surface are presented in Fig. 2, using boxplot representations. As expected (see Analysis of the intrinsic properties of the resolution matrix), all metrics values for MNE PSF and MNE CT were identical. Similarly, all metrics values for MNE CT, dSPM CT, and sLORETA CT were the same. Moreover DLE values for sLORETA PSF were correctly found to be zero.

The

Discussion

In this study, we applied the concept of resolution matrix to compare four source imaging methods, namely MNE, dSPM, sLORETA and cMEM on the basis of spatial accuracy measured with PSF and CT scores. The spatial properties of the imaging techniques were also analyzed by applying source localization of the N20 peak of the somatosensory evoked response to electrical stimulation of the median nerve.

Whereas most methods can reach overall good performance in accurately localizing the main peak of

Conclusion

We used the resolution matrix to compare different source imaging techniques (MNE, dSPM, sLORETA, and cMEM) in terms of PSF (related to the intrinsic spatial resolution of the source reconstruction) and CT, an important feature to be characterized before assessing functional connectivity between different regions. In this study combining theoretical analysis of resolution matrices and localization of real somatosensory data, we carefully evaluated both MEG and hdEEG data using similar amount of

Acknowledgments

This work was supported by NSERC Discovery grant (RGPIN/356610-2013) and CIHR (MOP-93614). TH was supported by the Irma Bauer Fellowship (McGill University, Faculty of Medicine). GP was supported by the Richard and Edith Strauss Canada Foundation. The authors would like to thank Ümit Aydin for his help in the correction of this manuscript.

References (58)

  • A. Molins et al.

    Quantification of the benefit from integrating MEG and EEG data in minimum l2-norm estimation

    NeuroImage

    (2008)
  • G.S. Russell et al.

    Geodesic photogrammetry for localizing sensor positions in dense-array EEG

    Clin. Neurophysiol.

    (2005)
  • H. Stefan

    Magnetic source imaging

    Rev. Neurol.

    (2009)
  • J. Vorwerk et al.

    A guideline for head volume conductor modeling in EEG and MEG

    NeuroImage

    (2014)
  • S.P. Ahlfors et al.

    Sensitivity of MEG and EEG to source orientation

    Brain Topogr.

    (2010)
  • S.P. Ahlfors et al.

    Cancellation of EEG and MEG signals generated by extended and distributed sources

    Hum. Brain Mapp.

    (2011)
  • Allison, T., McCarthy, G., Luby, M., Puce, A., Spencer, D.D., 1995. Localization of functional regions of human mesial...
  • C. Amblard et al.

    Biomagnetic source detection by maximum entropy and graphical models

    IEEE Trans. Bio-Med. Eng.

    (2004)
  • Ü. Aydin et al.

    Combining EEG and MEG for the reconstruction of epileptic activity using a calibrated realistic volume conductor model

    PLoS One

    (2014)
  • Chowdhury, R., Merlet, I., Birot, G., Kobayashi, E., Nica, A., Biraben, A., Wendling, F., Lina, J., Albera, L., Grova,...
  • R.A. Chowdhury et al.

    MEG source localization of spatially extended generators of epileptic activity: comparing Entropic and Hierarchical Bayesian approaches

    PLoS One

    (2013)
  • R.A. Chowdhury et al.

    MEG - EEG information fusion and electromagnetic source imaging: from theory to clinical application in epilepsy

    Brain Topogr.

    (2015)
  • Dale, A.M., Sereno, M.I., 1993. Improved localizadon of cortical activity by combining eeg and meg with mri cortical...
  • Darvas, F., Pantazis, D., Kucukaltun-Yildirim, E., Leahy, R.M., 2004. Mapping human brain function with MEG and EEG:...
  • J.S. Ebersole et al.

    Combining MEG and EEG source modeling in epilepsy evaluations

    J. Clin. Neurophysiol. : Off. Publ. Am. Electroencephalogr. Soc.

    (2010)
  • Eickhoff, S.B., Grefkes, C., Fink, G.R., Zilles, K., 2008. Functional lateralization of face, hand, and trunk...
  • A. Gramfort et al.

    OpenMEEG: opensource software for quasistatic bioelectromagnetics

    Biomed. Eng. Online

    (2010)
  • Grave de Peralta, R., Hauk, O., Gonzalez, S.L., Peralta, R.G.D., Hauk, O., Gonzalez, S.L., 2009. The...
  • C. Grova et al.

    Intracranial EEG potentials estimated from MEG sources: a new approach to correlate MEG and iEEG data in epilepsy

    Hum. Brain Mapp.

    (2016)
  • Cited by (95)

    View all citing articles on Scopus
    View full text