## MEG source localization under multiple constraints: an extended Bayesian framework (2006)

Venue: | NeuroImage |

Citations: | 28 - 4 self |

### BibTeX

@ARTICLE{Mattout06megsource,

author = {Jérémie Mattout and Christophe Phillips and B William D. Penny and Michael D. Rugg and Karl J. Friston A},

title = {MEG source localization under multiple constraints: an extended Bayesian framework},

journal = {NeuroImage},

year = {2006},

pages = {753--767}

}

### OpenURL

### Abstract

To use Electroencephalography (EEG) and Magnetoencephalography (MEG) as functional brain 3D imaging techniques, identifiable distributed source models are required. The reconstruction of EEG/ MEG sources rests on inverting these models and is ill-posed because the solution does not depend continuously on the data and there is no unique solution in the absence of prior information or constraints. We have described a general framework that can account for several priors in a common inverse solution. An empirical Bayesian framework based on hierarchical linear models was proposed for the analysis of

### Citations

993 | Bayes factors
- Kass, Raftery
- 1995
(Show Context)
Citation Context ...d their various combinations. Furthermore, Bayes factors B12 defined as the ratio of the evidences associated with model P1 and P2 enable us to compare two prior models when applied to the same data (=-=Kass and Raftery, 1995-=-) pðM=P1Þ B12 : ð9Þ pðM=P2Þ When B12 > 1, the data favor model P1 over model P2, and when B12 < 1, the data favor model P2. To make decisions based on Bayes factors, some cut-off value is required. ... |

834 |
V.: Solution of Ill-Posed Problems
- Tikhonov, Arsenin
- 1977
(Show Context)
Citation Context ...on the geometry and conductivity of head tissues (Mosher et al., 1999). Classical regularization: single prior A classical approach to the inverse problem is the weighted minimum norm (WMN) solution (=-=Tikhonov and Arsenin, 1977-=-). Simple and convenient, it has become a standard solution for the MEG inverse problem as stated by Eq. (1) (Hämäläinen and Llmoniemi, 1994; Dale and Halgren, 2001; Hauk, 2004). It entails minimizing... |

775 | A view of the em algorithm that justifies incremental, sparse, and other variants
- Neal, Hinton
- 1998
(Show Context)
Citation Context ...statistical thermodynamics and is given by the sum of potential and entropy terms Z Z FqJ ð ð Þ;k;lÞ qðJÞln pðJ;M=k;lÞdJqðJÞln qðJÞdJ; ð16Þ where q(J) is any distribution over the model parameters (=-=Neal and Hinton, 1998-=-).766 J. Mattout et al. / NeuroImage 30 (2006) 753–767 The log-likelihood is implicitly maximized through the maximization of F. This is achieved using an EM algorithm which alternates between maximi... |

275 |
Generalized cross-validation as a method for choosing a good ridge parameter
- Golub, Heath, et al.
- 1979
(Show Context)
Citation Context ...location priors is particularly relevant for multimodal fusion. Note that other approaches have been proposed to estimate hyperparameters. For instance, the Generalized Cross-Validation method (GCV) (=-=Golub et al., 1979-=-) could be extended to estimate several regularization parameters. However, contrary to the ReML approach, the GCV method does not provide any estimate of (multiple) noise components. Moreover, in the... |

182 |
Analysis of discrete ill-posed problems by means of the L-curve
- Hansen
- 1992
(Show Context)
Citation Context ...ch as derived from fMRI have also been considered (Liu et al., 1998; Phillips et al., 2002). The WMN solution depends upon the hyperparameter k, which is usually evaluated using the L-curve approach (=-=Hansen, 1992-=-; Gorodnitsky et al., 1995; Babiloni et al., 2001) (cf. Appendix A). A major drawback of this heuristic is that it cannot readily be extended to the estimation of multiple hyperparameters. Therefore, ... |

147 |
Maximum likelihood approaches to variance component estimation and to related problems
- Harville
- 1977
(Show Context)
Citation Context ...eme allows the simultaneous estimation of parameters J and hyperparameters (k,l) while accounting for the loss of degrees of freedom in the model incurred from estimating J when (k,l) are calculated (=-=Harville, 1977-=-). The ReML estimation consists of maximizing the likelihood of the observed data p(M/k,l), conditional on the hyperparameters, in the presence of parameters J. This is equivalent to maximizing the lo... |

136 | Improved localization of cortical activity by combining EEG and MEG with MRI cortical surface reconstruction: A linear approach
- Dale, Sereno
- 1993
(Show Context)
Citation Context ...address the anatomical deployment of an active region. In contradistinction to ECD approaches, a DM uses the subject’s anatomy derived from high resolution anatomical Magnetic Resonance Images (MRI) (=-=Dale and Sereno, 1993-=-). The solution space and associated forward models can then be made as realistic as allowed by computational constraints and the precision of head tissue conductivity measures. Moreover, due to the u... |

102 |
Basic mathematical and electromagnetic concepts of the biomagnetic inverse problem
- Sarvas
- 1987
(Show Context)
Citation Context ...on of cortical topology, since the mean distance between two neighboring dipoles was about 3 mm. We calculated the forward operator K, for this dipole mesh, using a single-shell spherical head model (=-=Sarvas, 1987-=-). Simulations MEG data were simulated over 130 sensors spread uniformly over the head, by activating two extended sources (cf. Fig. 1). Each source was a cluster comprising one randomly chosen dipole... |

99 | Classical and Bayesian inference in neuroimaging: Theory. NeuroImage
- Friston, Penny, et al.
- 2002
(Show Context)
Citation Context ...r source space. The optimal weight associated with each constraint is estimated from the data following an empirical Bayesian approach and is computed iteratively using Expectation Maximization (EM) (=-=Friston et al., 2002-=-). These weights are equivalent to restricted maximum likelihood (ReML) estimates of the prior covariance components. In a companion paper (Phillips et al., 2005), we addressed the face validity of em... |

95 |
Low resolution electromagnetic tomography: a new method for localizing electrical activity in the brain
- Pascual-Marqui, Michel, et al.
- 1994
(Show Context)
Citation Context ...matrix (Hämäläinen and Llmoniemi, 1994).When k tends to zero, one obtains the maximum likelihood (ML) solution. The wellknown LORETA approach entails setting Wp equal to a spatial Laplacian operator (=-=Pascual-Marqui et al., 1994-=-). This is also known as the maximum smoothness solution. Finally, external functional constraints such as derived from fMRI have also been considered (Liu et al., 1998; Phillips et al., 2002). The WM... |

93 |
Electromagnetic brain mapping
- Baillet, Mosher, et al.
- 2001
(Show Context)
Citation Context ...f neurons (Nunez and Silberstein, 2000). Unfortunately, localizing those electromagnetic sources is an ill-posed inverse problem that, in the absence of constraints, does not admit a unique solution (=-=Baillet et al., 2001-=-). Consequently, deriving a realistic and unique solution rests on prior knowledge, in addition to the observed measurements. Any source reconstruction approach is characterized by three components. T... |

89 |
From 3D magnetic resonance images to structural representations of the cortex topography using topology preserving deformations
- Mangin, Frouin, et al.
- 1995
(Show Context)
Citation Context ...etic Resonance Image (MRI) (Dale and Sereno, 1993). Following segmentation of the MRI volume, dipoles are typically located at each node of a triangular mesh covering the white/grey matter interface (=-=Mangin, 1995-=-). Furthermore, since the apical dendrites of these cortical neurons are organized perpendicularly to the surface, the corresponding dipoles are often constrained to this normal orientation. To simula... |

80 | Comparing dynamic causal models
- Penny, Stephan, et al.
- 2004
(Show Context)
Citation Context ...ell as their spatial extent. Finally, we introduced a supra ordinate level of inference. This was based on the log-evidence or data likelihood given a particular model (Trujillo-Barreto et al., 2004; =-=Penny et al., 2004-=-; Daunizeau et al., 2005). We used this probability-based criterion to identify the best prior model and, more generally, the best strategy for introducing priors. Unlike the LE and ROC criteria, the ... |

74 |
BNeuromagnetic Source Imaging with FOCUSS: A Recursive Weighted Minimum Norm Algorithm
- Gorodnitsky, George, et al.
- 1995
(Show Context)
Citation Context ... context of distributed approaches, priors based on mathematical, anatomical, physiological and functional heuristics have been considered (Hämäläinen and Llmoniemi, 1994; PascualMarqui et al., 1994; =-=Gorodnitsky et al., 1995-=-; Baillet and Garnero, 1997; Dale et al., 2000; Phillips et al., 2002; Mattout et al., 2003; Babiloni et al., 2004). Although these approaches involve different constraints and inverse criteria, they ... |

64 |
Eeg and meg: Forward solutions for inverse methods
- Mosher, Leahy, et al.
- 1999
(Show Context)
Citation Context ...ocation (Ermer et al., 2001). K is obtained by solving the forward problem for each dipole location and orientation of the given DM. It relies only upon the geometry and conductivity of head tissues (=-=Mosher et al., 1999-=-). Classical regularization: single prior A classical approach to the inverse problem is the weighted minimum norm (WMN) solution (Tikhonov and Arsenin, 1977). Simple and convenient, it has become a s... |

59 |
Interpreting magnetic fields of the brain: minimum norm estimates
- Hämäläinen, Ilmoniemi
- 1994
(Show Context)
Citation Context ...ext of selecting constraints (as opposed to sources). In the context of distributed approaches, priors based on mathematical, anatomical, physiological and functional heuristics have been considered (=-=Hämäläinen and Llmoniemi, 1994-=-; PascualMarqui et al., 1994; Gorodnitsky et al., 1995; Baillet and Garnero, 1997; Dale et al., 2000; Phillips et al., 2002; Mattout et al., 2003; Babiloni et al., 2004). Although these approaches inv... |

51 | A Bayesian approach to introducing anatomo-functional priors in the EEG/MEG inverse problem
- Baillet, Garnero
- 1997
(Show Context)
Citation Context ...ally, given a forward model, the third component is an inverse operator which, according to some criterion, defines a unique source distribution. For instance, when based on a probabilistic approach (=-=Baillet and Garnero, 1997-=-; Schmidt et al., 1999; Phillips et al., 2002; Amblard et al., 2004), the unique inverse solution corresponds to 1053-8119/$ - see front matter D 2005 Elsevier Inc. All rights reserved. doi:10.1016/j.... |

42 | Evoked dipole source potentials of the human auditory cortex. Electroencephalogr Clin Neurophysiol 65 - Scherg, Cramon - 1986 |

29 | An empirical Bayesian solution to the source reconstruction problem in EEG
- Phillips, Mattout, et al.
- 2005
(Show Context)
Citation Context ...ly using Expectation Maximization (EM) (Friston et al., 2002). These weights are equivalent to restricted maximum likelihood (ReML) estimates of the prior covariance components. In a companion paper (=-=Phillips et al., 2005-=-), we addressed the face validity of empirical Bayes in this context. In the present paper, the proposed framework is applied to simulated event related field (ERF) data with realistic noise. Our inve... |

28 |
Bayesian inference applied to the electromagnetic inverse problem
- Schmidt, George, et al.
- 1999
(Show Context)
Citation Context ..., the third component is an inverse operator which, according to some criterion, defines a unique source distribution. For instance, when based on a probabilistic approach (Baillet and Garnero, 1997; =-=Schmidt et al., 1999-=-; Phillips et al., 2002; Amblard et al., 2004), the unique inverse solution corresponds to 1053-8119/$ - see front matter D 2005 Elsevier Inc. All rights reserved. doi:10.1016/j.neuroimage.2005.10.037... |

27 |
Spatiotemporal mapping of brain activity by integration of multiple imaging modalities
- Dale, Halgren
- 2001
(Show Context)
Citation Context ...ed minimum norm (WMN) solution (Tikhonov and Arsenin, 1977). Simple and convenient, it has become a standard solution for the MEG inverse problem as stated by Eq. (1) (Hämäläinen and Llmoniemi, 1994; =-=Dale and Halgren, 2001-=-; Hauk, 2004). It entails minimizing the quadratic energy function UðJÞ jjM KJjj 2 þ kjjJjj2 Wn Wp ; where ||.||W 2 indicates the L2 norm associated with metric W and k is a hyperparameter which tun... |

14 |
Keep it simple: a case for using classical minimum norm estimation in the analysis of EEG and MEG data
- Hauk
- 2004
(Show Context)
Citation Context ...alanced way. Most can be framed in terms of a Weighted Minimum Norm criterion (WMN), which represents the classical and most popular distributed approach (see Classical regularization: single prior) (=-=Hauk, 2004-=-). However, a critical outstanding issue lies in the relative weighting of the accuracy and regularization criteria upon which the solution depends. Usually, in the context of Tikhonov regularization ... |

14 |
Statistical comparison of two ROC-curve estimates obtained from partially-paired datasets. Med. Decis. Making 18
- Metz, Herman, et al.
- 1998
(Show Context)
Citation Context ...lse positive rate (1 specificity). To assess the false positive rate, we chose 10 random dipoles among the nonactive ones. The area under the curve (AUC) quantifies the detection power of the method (=-=Metz, 1998-=-). The AUC ranges between 0 and 1 and indicates the probability of correct separation of an active source from a non active one. Comparing the AUC of different inverse models enables one to assess the... |

10 |
Rapidly recomputable eeg forward models for realistic head shapes
- J, Mosher, et al.
- 2001
(Show Context)
Citation Context ...cribe the measurements observed across sensors, induced by a particular dipole. The rows of K are the ‘‘lead fields’’ and describe the flow of current for a given sensor through each dipole location (=-=Ermer et al., 2001-=-). K is obtained by solving the forward problem for each dipole location and orientation of the given DM. It relies only upon the geometry and conductivity of head tissues (Mosher et al., 1999). Class... |

8 |
Trends in EEG source localization
- Koles
- 1998
(Show Context)
Citation Context ...the source parameters in order to explain the data as accurately as possible. In the iterative process, the source parameters are modified to minimize the residual error (Scherg and von Cramon, 1986; =-=Koles, 1998-=-). The solution is very sensitive to the number of sources and initial parameters (dipole locations and orientations), which need to be specified a priori. Indeed, ECD models require non-linear optimi... |

6 |
Biomagnetic source detection by maximum entropy and graphical models
- Amblard, Lapalme, et al.
- 2004
(Show Context)
Citation Context ...which, according to some criterion, defines a unique source distribution. For instance, when based on a probabilistic approach (Baillet and Garnero, 1997; Schmidt et al., 1999; Phillips et al., 2002; =-=Amblard et al., 2004-=-), the unique inverse solution corresponds to 1053-8119/$ - see front matter D 2005 Elsevier Inc. All rights reserved. doi:10.1016/j.neuroimage.2005.10.037754 J. Mattout et al. / NeuroImage 30 (2006)... |

6 |
Assessing the relevance of fMRI-based prior in the EEG inverse problem: a Bayesian model comparison approach
- Daunizeau, Grova, et al.
- 2005
(Show Context)
Citation Context ... extent. Finally, we introduced a supra ordinate level of inference. This was based on the log-evidence or data likelihood given a particular model (Trujillo-Barreto et al., 2004; Penny et al., 2004; =-=Daunizeau et al., 2005-=-). We used this probability-based criterion to identify the best prior model and, more generally, the best strategy for introducing priors. Unlike the LE and ROC criteria, the logevidence can be calcu... |

6 |
2004. Localization of realistic cortical activity in MEG using current multipoles
- Jerbi, Baillet, et al.
(Show Context)
Citation Context ...rse method can be distinguished by their respective source models: the equivalent current dipole (ECD) and the distributed model (DM). Although other source models have been used, such as multipoles (=-=Jerbi et al., 2004-=-) or continuous current densities (Riera et al., 1998), both approaches usually rely upon a dipolar representation of cortical sources, which are parameterized in terms of location, orientation and in... |

5 |
EEGdistributed inverse solutions for a spherical head model. Inverse Problems
- Riera, Fuentes, et al.
- 1009
(Show Context)
Citation Context ...ource models: the equivalent current dipole (ECD) and the distributed model (DM). Although other source models have been used, such as multipoles (Jerbi et al., 2004) or continuous current densities (=-=Riera et al., 1998-=-), both approaches usually rely upon a dipolar representation of cortical sources, which are parameterized in terms of location, orientation and intensity. An ECD models the activity of a large cortic... |

4 |
Improved realistic Laplacian estimate of highly-sampled EEG potentials with regularization techniques
- Babiloni, Carducci, et al.
- 1998
(Show Context)
Citation Context ... does not provide any estimate of (multiple) noise components. Moreover, in the context of a single hyperparameter, better results are sometimes obtained with the classical Tikhonov/L-curve approach (=-=Babiloni et al., 1998-=-). Finally, we demonstrated the effectiveness of the extended Bayesian framework for data-driven model selection and comparison. Based on model evidence, Bayesian model selection enables one to identi... |

4 |
Multimodal integration of EEG and MEG data: a simulation study with variable signal-to-noise ratio and number of sensors
- Babiloni, Babiloni, et al.
- 2004
(Show Context)
Citation Context ... have been considered (Hämäläinen and Llmoniemi, 1994; PascualMarqui et al., 1994; Gorodnitsky et al., 1995; Baillet and Garnero, 1997; Dale et al., 2000; Phillips et al., 2002; Mattout et al., 2003; =-=Babiloni et al., 2004-=-). Although these approaches involve different constraints and inverse criteria, they all obtain a unique solution by optimizing a goodness of fit term and a prior term in a carefully balanced way. Mo... |

4 |
The use of functional constraints for the neuroelectromagnetic inverse problem: alternatives and caveats
- Andino, Blanke, et al.
- 2001
(Show Context)
Citation Context ...these data fusion approach is the different time scale and nature of the physiological processes measured with MEG and fMRI techniques: as noted by many authors (Nunez and Silberstein, 2000; Gonzalez =-=Andino et al., 2001-=-; Pflieger and Greenblatt, 2001), an adaptative method that could distinguish between valid and invalid location priors is particularly relevant for multimodal fusion. Note that other approaches have ... |

4 |
Multivariate Source Prelocalization (MSP): use of functionally informed basis functions for better conditioning the MEG inverse problem
- Mattout, Pelegrini-Issac, et al.
- 2005
(Show Context)
Citation Context ..., this prior enforces correlation among neighboring sources. Intrinsic functional constraint It has been shown recently that an efficient regularization prior can be derived from the MEG data itself (=-=Mattout et al., 2005-=-). This is known as Multivariate Source Prelocalization (MSP) which provides, at each source location, a coefficient of activation. This coefficient refers to the spatial support of activity 2 Note th... |

4 |
Systematic regularisation of linear inverse solutions of the EEG source localisation problem
- Phillips, Rugg, et al.
- 2002
(Show Context)
Citation Context ...is an inverse operator which, according to some criterion, defines a unique source distribution. For instance, when based on a probabilistic approach (Baillet and Garnero, 1997; Schmidt et al., 1999; =-=Phillips et al., 2002-=-; Amblard et al., 2004), the unique inverse solution corresponds to 1053-8119/$ - see front matter D 2005 Elsevier Inc. All rights reserved. doi:10.1016/j.neuroimage.2005.10.037754 J. Mattout et al. ... |

3 |
Linear inverse source estimate of combined EEG and MEG data related to voluntary movements
- Babiloni, Carducci, et al.
- 2001
(Show Context)
Citation Context ...onsidered (Liu et al., 1998; Phillips et al., 2002). The WMN solution depends upon the hyperparameter k, which is usually evaluated using the L-curve approach (Hansen, 1992; Gorodnitsky et al., 1995; =-=Babiloni et al., 2001-=-) (cf. Appendix A). A major drawback of this heuristic is that it cannot readily be extended to the estimation of multiple hyperparameters. Therefore, only one single constraint can be considered. Thi... |

3 |
Localization Estimation Algorithm (LEA): a supervised prior based approach for solving the EEG/MEG inverse problem
- Mattout, Pélégrini-Issac, et al.
- 2003
(Show Context)
Citation Context ... functional heuristics have been considered (Hämäläinen and Llmoniemi, 1994; PascualMarqui et al., 1994; Gorodnitsky et al., 1995; Baillet and Garnero, 1997; Dale et al., 2000; Phillips et al., 2002; =-=Mattout et al., 2003-=-; Babiloni et al., 2004). Although these approaches involve different constraints and inverse criteria, they all obtain a unique solution by optimizing a goodness of fit term and a prior term in a car... |

3 |
Nonlinear analysis of multimodal dynamic brain imaging data
- Pflieger, Greenblatt
- 2001
(Show Context)
Citation Context ...roach is the different time scale and nature of the physiological processes measured with MEG and fMRI techniques: as noted by many authors (Nunez and Silberstein, 2000; Gonzalez Andino et al., 2001; =-=Pflieger and Greenblatt, 2001-=-), an adaptative method that could distinguish between valid and invalid location priors is particularly relevant for multimodal fusion. Note that other approaches have been proposed to estimate hyper... |

3 | Dynamic ReML solution for the EEG source reconstruction problem. Submitted abstract for HBM conference - Phillips, Mattout, et al. - 2004 |

3 |
Model selection in spatio-temporal electromagnetic sources analysis
- Waldorp, Huizenga, et al.
- 2005
(Show Context)
Citation Context ...specified a priori. Indeed, ECD models require non-linear optimization with the possibility of local minima. Moreover, determining the optimal number of ECD (model complexity) is a non-trivial issue (=-=Waldorp et al., 2005-=-); some simulation studies have shown that, even with the right number of sources, ECD approaches are less reliable than distributed ones, when dealing with more than one source (Yao and Dewald, 2005)... |

2 |
Dynamic statistical parametric This algorithm id implemented (spm_reml.m) in the current version of the SPM software (Wellcome Department of Imaging Neuroscience, http:// www.fil.ion.ucl.ac.uk/spm (2002)). mapping: combining fMRI and MEG for high-resoluti
- Dale, Liu, et al.
- 2000
(Show Context)
Citation Context ...athematical, anatomical, physiological and functional heuristics have been considered (Hämäläinen and Llmoniemi, 1994; PascualMarqui et al., 1994; Gorodnitsky et al., 1995; Baillet and Garnero, 1997; =-=Dale et al., 2000-=-; Phillips et al., 2002; Mattout et al., 2003; Babiloni et al., 2004). Although these approaches involve different constraints and inverse criteria, they all obtain a unique solution by optimizing a g... |

2 |
Spatiotemporal imaging of human activity using functional MRI constrained magnetoencephalography data: Monte Carlo simulations
- Liu, Belliveau, et al.
- 1998
(Show Context)
Citation Context ... Laplacian operator (Pascual-Marqui et al., 1994). This is also known as the maximum smoothness solution. Finally, external functional constraints such as derived from fMRI have also been considered (=-=Liu et al., 1998-=-; Phillips et al., 2002). The WMN solution depends upon the hyperparameter k, which is usually evaluated using the L-curve approach (Hansen, 1992; Gorodnitsky et al., 1995; Babiloni et al., 2001) (cf.... |