The localization approach to MEG source estimation considers that brain activity at any time instant is generated by a relatively small number (a handful, at most) of brain regions. Each source is therefore represented by an elementary model, such as an ECD, that captures local distributions of neural currents. Ultimately, each elementary source is back projected or constrained to the subject’s brain volume or an MRI anatomical template, for further interpretation. Localization models are essentially compact, in terms of number of generators involved and their surface extension (from point-like to small cortical surface patches).

The early MEG literature is abundant in studies reporting on single-dipole source models. The somatotopic, tonotopic, and retinotopic responses are examples where the single-dipole model contributed to the better temporal characterization of primary sensory area responses. However, later components of evoked fields usually necessitate that more elementary sources be adjusted, and this is detrimental to the numerical stability and robustness of the inverse model. The number of elementary sources in the model is often qualitatively assessed by expert users, which may undermine the reproducibility of the analyses. Hence, special care should be brought to the evaluation of the stability and robustness of the estimated source models. With all that in mind, source localization techniques have proven to be effective, even in complex experimental paradigms (Helenius, Parviainen, Paetau, & Salmelin, 2009). Indeed, as we outline below, dipole-based approaches currently constitute the most widely applied and best-replicated paradigms for presurgical language mapping with MEG. The alternative imaging approaches to source modeling were originally inspired by the plethoric research in digital image restoration and reconstruction in other fields of engineering.

In these approaches, dense grids of current dipoles are predefined over the entire brain volume or limited to the cortical gray matter surface. These dipoles are fixed in location and, generally, orientation, and their amplitudes are estimated at once. Thus, the nonlinear parameters (e.g., the elementary source locations) now become fixed priors as provided by anatomical information, and the remaining free parameters are the amplitudes of the elementary source currents distributed on the brain’s geometry. The resulting image source models do not yield small sets of local elementary models but rather the distribution of “all” neural currents. As the estimated sources are spatially fixed in this approach, they are homologous to the pixels of a digital image. Hence, the resulting estimations are a set of images or maps, representing an estimation of the intensity of neural currents, distributed within the entire brain volume or constrained at the cortical surface. Contrary to the localization model though, there is no intrinsic sense of distinct, active source regions per se. Explicit identification of activity issued from discrete brain regions usually necessitates complementary analysis, such as empirical or inference-driven amplitude thresholding, to discard elementary sources of nonsignificant contribution according to the statistical appraisal. In that respect, MEG source images are very similar in essence to the activation maps obtained in fMRI, with the supplementary benefit of time resolution (Fig. 9.2).

Just like in the context of source localization where, for example, the number of sources is a restrictive prior as a remedy to ill-posedness, imaging models also must be complemented by a priori information. Several academic software packages have addressed this issue by implementing variations of the well-described minimum norm model (Lin, et al., 2006; Lin, Belliveau, Dale, & Hämäläinen, 2006). For present purposes, it is important only that readers be aware of the equivalent need for prior information in both the localization and imaging approaches. These priors are necessary to determine a workable solution to a problem with an infinite number of solutions, in theory.

A final set of approaches involve signal classification and spatial filtering (e.g., “beam forming”) to translate the source localization problem into a signal detection issue. In essence these are scanning techniques that systematically sift through the brain space to evaluate how a predetermined elementary source model would fit the data at every voxel of the brain volume. These techniques are efficient alternative approaches in that respect and have gained considerable momentum in the MEG community in the recent years. They have proven especially fruitful in several applications (for both source estimation and artifact suppression) and the interested reader is referred to a review article by Hillebrand, Singh, Holliday, Furlong, and Barnes (2005) for an introduction to the approach.

Each of these three approaches, MEG source localization, imaging, and scanning, makes use of phenomenally complex methodology and modeling techniques to noninvasively localize the source of magnetic neural activity within the brain. As with any experimental observation, they are subject to error and benefit from explicit assessments of their sensitivity to measure noise and modeling approximations (see, e.g., Darvas, et al., 2005; Mosher, Spencer, Leahy, & Lewis, 1993). Ultimately, an informed consideration of their various assumptions, strengths, and weaknesses, as described above, can guide their optimal application to any given clinical evaluation.

Despite the many complex aspects of magnetic source imaging, the preceding exposition can easily be reduced to a handful of essential points to consider before beginning a program of MEG presurgical evaluation. As with fMRI, an efficiently functioning MEG laboratory requires the effort of a multidisciplinary team. The various requirements would ideally be met by a group of individuals with collective expertise in MEG physics, data acquisition, software engineering, task design and development, cognitive evaluation, and neuroanatomy, epileptology, neuro-oncology, etc.

Developing a standardized data acquisition and analysis protocol simply requires explicitly planning through the steps outlined above (e.g., subject preparation, data acquisition, quality control, source modeling, etc). It is important to remember that the tasks, stimuli, and analysis approach in MEG protocols should be designed in a coordinated and complementary fashion and specifically with sensitivity to how the temporal aspects of the stimuli, and expected behavioral and neural responses will condition the access to a subset of methods for data analysis. For example, it might be discussed whether or not to collect behavioral data simultaneously with MEG or whether the data should be analyzed as stimulus-locked vs. response-locked epochs. Event-related responses in cognitive tasks most often contain both early components (50–200 ms post-stimulus) reflecting the primary sensory processing and integration of the stimulus and later components (>200 ms post-stimulus) that are typically thought to reflect activity in association cortices. It has therefore been argued that the mapping of higher cognitive functions with MEG may require restricting analyses to the later components of the brain responses (Simos, Papanicolaou, Castillo, & Buchanan, 2009). Decisions such as duration of stimulus presentation and length and variability of interstimulus intervals (ISIs) need to be guided by the specifics of the task modality and of the underlying scientific hypothesis to be tested (e.g., visual stimuli tend to elicit saccades, auditory stimuli unfold over time, etc.) and the expected nature of the neural responses of interest (Salmelin & Parkkonen, 2010). There are several academic or commercial software packages for MEG data analysis, and the major MEG vendors provide their own software for many aspects of data processing. Additionally, the reference papers for several excellent and freely available software packages can be found at: http://​www.​hindawi.​com/​journals/​cin/​2011/​si.​ebm/​.

As discussed above, MEG recording is greatly facilitated if caution is taken prior to scanning to remove easily eliminated sources of magnetic artifacts (e.g., pieces of metal in garments). To this end, it is a good investment of time to briefly conduct a noise run with the subject sitting or laying under the MEG dewar to help identify any unsuspected artifacts, before subject preparation is completed by applying HPI coils and, optionally, EEG leads. This step has the additional benefit of familiarizing the patient with the MSR environment before the clinical scan per se. For the most part, the prior recommendations regarding behavioral monitoring and testing outside of the instrument in the context of fMRI also apply to MEG. Also like fMRI, it has been reported that language-related activity is impacted by factors such as subject alertness, task compliance, and IQ (Merrifield, Simos, Papanicolaou, Philpott, & Sutherling, 2007; Simos et al., 2009), though there are ongoing efforts to develop MEG language protocols that are better suited for use with pediatric and cognitively lower-functioning individuals. Unlike the highly magnetized MRI environment, patient safety concerns within the MSR primarily relate to clinical concerns. The most important of these is always having appropriately trained personnel ready to rapidly enter and assist the patient within the enclosed MSR, in case of an acute seizure or other medical emergency.

For the same reasons outlined in the fMRI chapter of this volume, neuropsychologists can clearly play an appropriate and valuable role in the design, administration, and interpretation of MEG studies of higher cognitive functions. In 2009, the American Academy of Neurology published a policy regarding MEG, including indications to “localize and preserve eloquent cortex during resective surgery,” in the context of surgeries for tumors and arteriovenous malformations, in addition to MEG indications for epilepsy surgical planning (see Wu et al., 2010). MEG CPT codes have been issued for these clinical conditions http://​acmegs.​org/​id77.​html under “MEG recording and analysis” for localization of language, as well as sensory and motor functions, with the former representing an excellent venue for integrating neuropsychological expertise. As of this writing, we are not aware of any official position statements regarding the role of neuropsychologists in the clinical use of MEG. Given the historical importance that localization of cognitive functions played within the discipline of clinical neuropsychology, as well as ongoing discussions about expanding roles for neuropsychologists (Bilder, 2011; Miller, Elbert, Sutton, & Heller, 2007), this may be a timely topic for discussion within neuropsychological professional organizations and training programs.

There are a number of tasks that have been utilized to map language-specific cortical regions with MEG. Fewer studies have examined episodic memory with MEG in ways that are immediately relevant to presurgical epilepsy evaluations. By far, the most widely studied and best-replicated task for evaluation of the laterality of language functions with MEG is the continuous recognition memory (CRM) paradigm developed by Papanicolaou and colleagues (Frye, Rezaie, & Papanicolaou, 2009). A common version of the task involves presenting the subject with a list of target words (abstract English nouns) immediately prior to the scan, which they are asked to study and accurately repeat. The acquisition scan involves presenting the subject with frequent spoken targets that are randomly interspersed with novel distractor items. Subjects are instructed to immediately issue a manual response (a finger lift) upon presentation of a target and otherwise make no response. Thus, the task essentially represents an auditory verbal recognition memory paradigm.

The analysis stream most often utilized for this task involves identifying the locations of successive dipolar sources that account for the magnetic flux distribution in the “late” (i.e., >200 ms) portion of the post-stimulus period. The acquisition run is often divided into two subsets where the spatial and temporal overlap of the obtained dipoles is assessed, and those which are reliable between the two subsets are retained. The resulting dipole maps depict clusters of reliable source locations. In a manner similar to fMRI, a laterality index can be calculated by taking the difference between the number of right- and left-hemisphere sources divided by their sum, thereby providing a hemispheric dominance index (LI) that varies between 1 and −1 (see Simos et al., 2009 for additional details). This task has been shown to localize activity within posterior language areas (i.e., in the vicinity of the superior and middle temporal gyri) in a number of empirical reports (e.g., Breier et al., 2001; Merrifield et al., 2007; Papanicolaou et al., 1999) including large samples of epilepsy patients (Papanicolaou et al., 2004), in both auditory and visual stimulus modalities (Papanicolaou et al., 2006) and in samples of non-English speakers.

As recently reviewed by Pirmoradi, Béland, Nguyen, Bacon, and Lassonde (2010), many other tasks have also been investigated for functionally mapping language functions with MEG. In addition to variants of the CRM task outlined above, they cite numerous other paradigms designed to either determine hemispheric language dominance or localize systems related to specific functions (Pirmoradi et al., 2010). These include: passive listening and counting paradigms (Kim & Chung, 2008; Martin et al., 1993; Szymanski et al., 2001; Szymanski, Rowley, & Roberts, 1999); grammatical or semantic categorization, matching, or judgment (Härle, Dobe, Cohen, & Rockstroh, 2002; Kamada et al., 2006, 2007; McDonald et al., 2009; Simos, Breier, Zouridakis, & Papanicolaou, 1998); and note/tone vs. vowel comparisons (Gootjes, Raij, Salmelin, & Hari, 1999; Kirveskari, Salmelin, & Hari, 2006).

Other reports have more explicitly examined neural systems involved in language output, via both isolated and combined studies of covert reading and picture naming (Breier et al., 2001; Hirata et al., 2010; Kober et al., 2001), delayed and overt naming (Fisher et al., 2008; Levelt, Praamstra, Meyer, Helenius, & Salmelin, 1998; Salmelin, Hari, Lounasmaa, & Sams, 1994) and reading (Salmelin, Schnitzler, Schmitz, & Freund, 2000), and delayed, covert, and overt word or verb generation tasks (Bowyer et al., 2005; Fisher et al., 2008; Pang, Wang, Malone, Kadis, & Donner, 2011; Ressel et al., 2006; Yamamoto et al., 2006). Despite over a decade of research, and many reliable findings, the MEG language and memory mapping literature is still at somewhat of an earlier stage of development than that of fMRI, owing largely to the smaller MEG community and the more recent maturation of well-delineated source modeling approaches and the availability of software tools.

There have been relatively few MEG studies that have specifically examined memory functions in epilepsy. One study utilized a variant of the continuous recognition memory paradigm to examine medial temporal lobe activity in a group of healthy subjects and patients with left mesial temporal sclerosis (LMTS) (Ver Hoef, Sawrie, Killen, & Knowlton, 2008). The authors reasoned that since the performance of the CRM task involves both language and memory processes, the paradigm might elicit lateralized medial temporal lobe activity in addition to the robust activation it elicits in posterior temporal areas. A variant of the auditory CRM protocol was investigated in a final sample of six patients with LMTS and seven healthy controls, and a source localization approach involving clusters of dipoles was utilized. Across two runs of the CRM task, left mesial temporal dipole clusters were observed in the majority of control subjects but not LMTS patients (86 vs. 33 %). Conversely, right medial temporal dipole clusters were observed in all LMTS patients (100 %), relative to only 2 of 7 control subjects (28 %), with the difference being statistically significant. The direction of medial temporal LIs also significantly differed between groups, with patients demonstrating rightward mesial temporal laterality, relative to left lateralization in controls. Finally, patients and controls also significantly differed with regard to the number of runs on which each respective group showed right medial temporal activity (67 vs. 7 %).