Source Localization Methodology
Note: Reprints are for non-commercial, scholarly use
only. Please do not re-distribute them. You may have to login to access some copyrighted materials.
Web-Based Expertise
Reviews
Cross-Validation of Source-Localized EEG/MEG Data
fMRI
PET
Epileptic Foci (surgical, fMRI, PET, and intra-cerebral recording cross-validation)
Bagshaw et al., 2005 (fMRI)
Seeck et al., 1998 (intra-cerebral recordings)
Trebuchon-Da Fonseca et al., 2005 (intra-cerebral recordings)
Zumsteg et al., 2005a (intra-cerebral recordings)
Simulations
LORETA vs. sLORETA
LORETA vs. SPM
LORETA vs. Dipole Localization
LORETA vs. Various Methods: Simulated and Experimental Data
Potential Pitfalls of Dipole ("Deterministic") Source Localization Analyses
Statistical Analysis of Source Localized Activity
see my Quantitative Methods page
DCM Approach
Some Highly Personal Reflections on the Utility of Source Localization in the "Age of Functional MRI"
Source Localization is Employed in Reports Appearing in Top-Tier Journals
Numerous reports employing source localization have appeared in Science (Gehring & Willoughby, 2002; Makeig et al., 2002; Massimini et al., 2005), Nature Neuroscience (Koelsch et al., 2004; Maess et al., 2001; McDonald et al., 2005; Schneider et al., 2002; van Schie et al., 2004), PNAS (Gross et al., 2001; Jaaskelainen et al., 2004; Schoenfeld et al., 2003), and Neuron (Gonsalves et al., 2005; Noesselt et al., 2002) in recent years.
The Validity and Accuracy of Source Localization
Work using simulated data (Cuffin et al., 2001a, 2001b; Yao & Dewald, 2005) demonstrates that source localization techniques are able to localize activation to within 10-20mm (or 1-3 LORETA voxels) even for deep, mesial sources. Cross-validation studies of epileptic foci localization by the Zurich group (e.g., Lantz et al., 2003; Michel et al., 2004) provide similar estimates using high-density recordings. Work using PET and LORETA to independently localize epileptic loci also show high correspondence across modalities (Zumsteg et al., 2005b). Further corroboration comes from work showing substantial similarity between intra-cerebral recordings and source localization foci (Seeck et al., 1998; Trebuchon-Da Fonseca et al., 2005), including LORETA analyses of deep mesial temporal/hippocampal and subcallosal cingulate foci (Zumsteg et al., 2005a). Moreover, in two recent EEG/fMRI studies LORETA localizations were, on average 16 mm (Mulert et al., 2004) and 14.5 mm (Vitacco et al., 2002) from fMRI activation loci?including loci buried in the depths of the insula. Similar results have been reported by the Montreal group (Benar et al., 2005). Finally, work by our group published in Molecular Psychiatry (Pizzagalli et al., 2004) demonstrates excellent correspondence between LORETA measures of activation in subgenual cingulate (a sub-component of vmPFC) and PET measures of glucose metabolism.
Why Would Anyone Elect to Use Source Localization in the "Age of fMRI"?
Mainstream high field-strength EPI BOLD fMRI techniques suffer from four crucial limitations that render them questionable choices for investigating a variety of phenomena.
Signal dropout. First, owing to inhomogeneities in the static B0 field in ventromedial prefrontal cortex (vmPFC) arising from its proximity to the air-filled sinuses, mainstream fMRI techniques are incapable of measuring vmPFC (for a dramatic example of our laboratory?s experience with this phenomenon, see Nitschke et al., 2004, Figure 3), although special techniques have been recently developed to compensate for this limitation (e.g., Cusack et al., 2005; Preston et al., 2004). Such techniques are still not yet widely available. By contrast, source localization of high-density EEG can provide adequate measures of vmPFC.
Insensitivity to slow variations in experimental factors. Second, BOLD fMRI lacks a true quantitative baseline and is insensitive to experimental manipulations that vary more slowly than ~10-60 sec (0.10 - .017Hz; Aguirre & D?Esposito, 1999; Detre & Wang, 2002), such as emotion manipulations or drug administration. In fact, in a recent review of the potential utility of perfusion fMRI techniques under development, Aguirre et al. (2002) noted that, "Indeed, one could drive the notion of a "low-frequency design" to its logical limit and study a subject in one experimental condition during one 6-min scan and then another experimental condition during another 6-min scan...Such a design would have negligible power using BOLD fMRI." Unfortunately, perfusion fMRI techniques are currently employed at only a handful of research centers worldwide. Although PET techniques, which like perfusion fMRI possess a quantitative baseline, could be employed, they are even more expensive than fMRI and require the administration of radioactive tracers. Source localized EEG, on the other hand, has a quantitative baseline (amperes / meter-squared) and is well suited to measuring slowly unfolding changes in activation, such as that induced by manipulations of mood.
High cost makes larger samples impractical for routine investigations. Third, the low-cost of source-localized EEG makes it tractable to collect the relatively large sample necessitated by multivariate individual differences analyses. By contrast, fMRI studies of emotion typically have small samples (~10-25 participants), limiting the kinds of questions (e.g., statistical mediation) that can reasonably be asked.
Insensitivity to rapid changes in activation. Finally, as is widely appreciated, event-locked electrophysiological methods (i.e., event-related potentials in the time-domain and event-related spectral perturbations in the spectral-domain) provide millisecond temporal resolution. By contrast, mainstream fMRI techniques have a resolution on the order of about a second.