Limits...
Non-Gaussian probabilistic MEG source localisation based on kernel density estimation.

Mohseni HR, Kringelbach ML, Woolrich MW, Baker A, Aziz TZ, Probert-Smith P - Neuroimage (2013)

Bottom Line: There is strong evidence to suggest that data recorded from magnetoencephalography (MEG) follows a non-Gaussian distribution.By providing a Bayesian formulation for MEG source localisation, we show that the source probability density function (pdf), which is not necessarily Gaussian, can be estimated using multivariate kernel density estimators.The proposed non-Gaussian source localisation approach is shown to give better spatial estimates than the LCMV beamformer, both in simulations incorporating non-Gaussian signals, and in real MEG measurements of auditory and visual evoked responses, where the highly correlated sources are known to be difficult to estimate.

View Article: PubMed Central - PubMed

Affiliation: Institute of Biomedical Engineering, School of Engineering Science, University of Oxford, Oxford, UK; Department of Psychiatry, University of Oxford, Warneford Hospital, UK.

Show MeSH
Reconstruction of neural activity in the auditory paradigm using LCMV beamformer with (a) λ = 0.01%, (b) λ = 0.1% and (c) λ = 1% of trace of data covariance matrix, and using the proposed non-Gaussian method with (d) h = 1, (e) h = 5 and (f) h = 10.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4273612&req=5

f0055: Reconstruction of neural activity in the auditory paradigm using LCMV beamformer with (a) λ = 0.01%, (b) λ = 0.1% and (c) λ = 1% of trace of data covariance matrix, and using the proposed non-Gaussian method with (d) h = 1, (e) h = 5 and (f) h = 10.

Mentions: The results of the source reconstruction are presented in Fig. 11using the non-Gaussian PD and LCMV beamformers. In all the figures, the small green volume is the mask representing the primary auditory cortex presented using the Juelich Histological Atlas (Morosan et al., 2001). Figs. 11(a)–(c) show the results using the LCMV beamformer (blue regions) with λ equal to 0.01%, 0.1% and 1% of the trace. It is clear that, depending on the value of λ, the LCMV beamformer has only reconstructed a source primarily in the left or right auditory cortex, leaving out the source in the other cortex. In addition, it has placed an extraneous source in the middle of the brain (close to the precuneus and posterior cingulate cortex), and one deep in the brain. These extraneous sources are likely to arise from correlations between genuine sources. Figs. 11(d)–(f) show the results from the non-Gaussian method for h = 1, 5 and 10, respectively. It shows better reconstruction of both left and right sources in the primary auditory cortex (especially for h = 5). It does not show any source deep in the brain, but for higher values of h it does present a potentially incorrect source in the middle of the brain due to the high correlation between sources. These observations also show that the sensitivity of the new method to the choice of bandwidth h was no higher than the sensitivity of the LCMV beamformer to regularisation factor λ.


Non-Gaussian probabilistic MEG source localisation based on kernel density estimation.

Mohseni HR, Kringelbach ML, Woolrich MW, Baker A, Aziz TZ, Probert-Smith P - Neuroimage (2013)

Reconstruction of neural activity in the auditory paradigm using LCMV beamformer with (a) λ = 0.01%, (b) λ = 0.1% and (c) λ = 1% of trace of data covariance matrix, and using the proposed non-Gaussian method with (d) h = 1, (e) h = 5 and (f) h = 10.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4273612&req=5

f0055: Reconstruction of neural activity in the auditory paradigm using LCMV beamformer with (a) λ = 0.01%, (b) λ = 0.1% and (c) λ = 1% of trace of data covariance matrix, and using the proposed non-Gaussian method with (d) h = 1, (e) h = 5 and (f) h = 10.
Mentions: The results of the source reconstruction are presented in Fig. 11using the non-Gaussian PD and LCMV beamformers. In all the figures, the small green volume is the mask representing the primary auditory cortex presented using the Juelich Histological Atlas (Morosan et al., 2001). Figs. 11(a)–(c) show the results using the LCMV beamformer (blue regions) with λ equal to 0.01%, 0.1% and 1% of the trace. It is clear that, depending on the value of λ, the LCMV beamformer has only reconstructed a source primarily in the left or right auditory cortex, leaving out the source in the other cortex. In addition, it has placed an extraneous source in the middle of the brain (close to the precuneus and posterior cingulate cortex), and one deep in the brain. These extraneous sources are likely to arise from correlations between genuine sources. Figs. 11(d)–(f) show the results from the non-Gaussian method for h = 1, 5 and 10, respectively. It shows better reconstruction of both left and right sources in the primary auditory cortex (especially for h = 5). It does not show any source deep in the brain, but for higher values of h it does present a potentially incorrect source in the middle of the brain due to the high correlation between sources. These observations also show that the sensitivity of the new method to the choice of bandwidth h was no higher than the sensitivity of the LCMV beamformer to regularisation factor λ.

Bottom Line: There is strong evidence to suggest that data recorded from magnetoencephalography (MEG) follows a non-Gaussian distribution.By providing a Bayesian formulation for MEG source localisation, we show that the source probability density function (pdf), which is not necessarily Gaussian, can be estimated using multivariate kernel density estimators.The proposed non-Gaussian source localisation approach is shown to give better spatial estimates than the LCMV beamformer, both in simulations incorporating non-Gaussian signals, and in real MEG measurements of auditory and visual evoked responses, where the highly correlated sources are known to be difficult to estimate.

View Article: PubMed Central - PubMed

Affiliation: Institute of Biomedical Engineering, School of Engineering Science, University of Oxford, Oxford, UK; Department of Psychiatry, University of Oxford, Warneford Hospital, UK.

Show MeSH