Automated model selection in covariance estimation and spatial whitening of MEG and EEG signals

Engemann, Denis A.; Gramfort, Alexandre

doi:10.1016/j.neuroimage.2014.12.040

Cited by 176 publications

(149 citation statements)

References 38 publications

Supporting

Mentioning

147

Contrasting

Order By: Relevance

“…The SPM-faces MEG dataset 1 was recorded with a VSM MedTech Inc. whole-head axial gradiometer system with 274 channels (CTF/VSM) at the Functional Imaging Laboratory, London. The paradigm is detailed in [8].…”

Section: A Datasetsmentioning

confidence: 99%

“…The automated approach to covariance estimation is detailed in [1]. Here we used its implementation that is provided by the MNE software [9].…”

Section: B Covariance Estimationmentioning

confidence: 99%

“…The best and the worst estimates are obtained out of a series of alternative strategies using cross-validation and log-likelihood computation. The methods compared were Factor Analysis (FA), Shrunk Covariance (SC) diagonal regularization, and unregularized, empirical covariance [1].…”

Section: B Covariance Estimationmentioning

confidence: 99%

“…To image brain activity, sensor signals are projected onto a cortical model using inverse solvers. These solvers typically involve Euclidean L2 constraints which inherently assume Gaussian noise with zero mean and equal variances across sensors and therefore necessitate a spatial whitening step for input data [1]. The success of source localization thus depends on the quality of the covariance estimate from which the whitening is computed.…”

Section: Introductionmentioning

confidence: 99%

“…The most common approach to regularize the covariance is to down-weight off-diagonal matrix coefficients, while more advanced regularization methods are based on shrinkage techniques [2] or generative models with low rank This work was supported by the ANR grant THALAMEEG, ANR-14-NEUC-0002-01, and the German Federal Ministry of Education and Research (BMBF) (3IPT605A) assumptions: probabilistic PCA (PPCA) and factor analysis (FA) [3], [4]. Through the use of cross-validation, all of these models can be tuned and compared based on Gaussian likelihood computed on unseen data [1].…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Mind the Noise Covariance When Localizing Brain Sources with M/EEG

Engemann

Strohmeier²,

Larson

et al. 2015

2015 International Workshop on Pattern Recognition in NeuroImaging

Self Cite

View full text Add to dashboard Cite

Abstract-Magnetoencephalography (MEG) and electroencephalography (EEG) are imaging methods that measure neuronal dynamics non invasively with high temporal precision. It is often desired in MEG and EEG analysis to estimate the neural sources of the signals. Strategies used for this purpose often take into account the covariance between sensors to yield more precise estimates of the sources. Here we investigate in greater detail how the quality of such covariance estimates conditions the estimation of MEG and EEG sources. We investigated three distinct source localization methods: dynamic Statistical Parametric Maps (dSPM), the linearly constrained minimum variance (LCMV) beamformer and Mixed-Norm Estimates (MxNE). We implemented and evaluated automated strategies for improving the quality of covariance estimates at different stages of data processing. Our results show that irrespective of the source localization method, accuracy can suffer from improper covariance estimation but can be improved by relying on automated regularization of covariance estimates.

show abstract

Section: A Datasetsmentioning

confidence: 99%

“…The automated approach to covariance estimation is detailed in [1]. Here we used its implementation that is provided by the MNE software [9].…”

Section: B Covariance Estimationmentioning

confidence: 99%

Section: B Covariance Estimationmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Mind the Noise Covariance When Localizing Brain Sources with M/EEG

Engemann

Strohmeier²,

Larson

et al. 2015

2015 International Workshop on Pattern Recognition in NeuroImaging

Self Cite

View full text Add to dashboard Cite

show abstract

A statistical framework for neuroimaging data analysis based on mutual information estimated via a gaussian copula

Ince

Giordano

Kayser

et al. 2016

Human Brain Mapping

265

298

View full text Add to dashboard Cite

We begin by reviewing the statistical framework of information theory as applicable to neuroimaging data analysis. A major factor hindering wider adoption of this framework in neuroimaging is the difficulty of estimating information theoretic quantities in practice. We present a novel estimation technique that combines the statistical theory of copulas with the closed form solution for the entropy of Gaussian variables. This results in a general, computationally efficient, flexible, and robust multivariate statistical framework that provides effect sizes on a common meaningful scale, allows for unified treatment of discrete, continuous, unidimensional and multidimensional variables, and enables direct comparisons of representations from behavioral and brain responses across any recording modality. We validate the use of this estimate as a statistical test within a neuroimaging context, considering both discrete stimulus classes and continuous stimulus features. We also present examples of analyses facilitated by these developments, including application of multivariate analyses to MEG planar magnetic field gradients, and pairwise temporal interactions in evoked EEG responses. We show the benefit of considering the instantaneous temporal derivative together with the raw values of M/EEG signals as a multivariate response, how we can separately quantify modulations of amplitude and direction for vector quantities, and how we can measure the emergence of novel information over time in evoked responses. Open‐source Matlab and Python code implementing the new methods accompanies this article. Hum Brain Mapp 38:1541–1573, 2017. © 2016 Wiley Periodicals, Inc.

show abstract

Data‐driven model optimization for optically pumped magnetometer sensor arrays

Duque-Muñoz

Tierney

Meyer

et al. 2019

Human Brain Mapping

View full text Add to dashboard Cite

Optically pumped magnetometers (OPMs) have reached sensitivity levels that make them viable portable alternatives to traditional superconducting technology for magnetoencephalography (MEG). OPMs do not require cryogenic cooling and can therefore be placed directly on the scalp surface. Unlike cryogenic systems, based on a well‐characterised fixed arrays essentially linear in applied flux, OPM devices, based on different physical principles, present new modelling challenges. Here, we outline an empirical Bayesian framework that can be used to compare between and optimise sensor arrays. We perturb the sensor geometry (via simulation) and with analytic model comparison methods estimate the true sensor geometry. The width of these perturbation curves allows us to compare different MEG systems. We test this technique using simulated and real data from SQUID and OPM recordings using head‐casts and scanner‐casts. Finally, we show that given knowledge of underlying brain anatomy, it is possible to estimate the true sensor geometry from the OPM data themselves using a model comparison framework. This implies that the requirement for accurate knowledge of the sensor positions and orientations a priori may be relaxed. As this procedure uses the cortical manifold as spatial support there is no co‐registration procedure or reliance on scalp landmarks.

show abstract

Automated model selection in covariance estimation and spatial whitening of MEG and EEG signals

Cited by 176 publications

References 38 publications

Mind the Noise Covariance When Localizing Brain Sources with M/EEG

Mind the Noise Covariance When Localizing Brain Sources with M/EEG

A statistical framework for neuroimaging data analysis based on mutual information estimated via a gaussian copula

Data‐driven model optimization for optically pumped magnetometer sensor arrays

Contact Info

Product

Resources

About