|Home | About | Journals | Submit | Contact Us | Français|
The goal of the current study was to examine the dynamics of language lateralization using magnetoencephalographic (MEG) imaging, to determine the sensitivity and specificity of MEG-imaging, and to determine if MEG-imaging can become a viable alternative to the intracarotid amobarbital procedure (IAP), the current gold-standard for preoperative language lateralization in neurosurgical candidates.
MEG was recorded during an auditory verb-generation task and imaging analysis of oscillatory activity was initially performed in 21 subjects with epilepsy, brain tumor, or arteriovenous malformation who had undergone IAP and MEG. Time-windows and brain regions-of-interest that best discriminated between IAP determined left or right dominant for language were identified. Parameters derived in the retrospective analysis, was applied to a prospective cohort of 14 patients and healthy controls.
Power decreases in the beta-frequency band were consistently observed following auditory stimulation in inferior frontal, superior temporal, and parietal cortices; similar power decreases were also seen in inferior frontal cortex prior to and during overt verb generation. Language lateralization was clearly observed to be a dynamic process that is bilateral for several hundred milliseconds during periods of auditory perception and overt speech production. Correlation with the IAP was seen in 13 of 14 (93%) of prospective patients, with the test demonstrating a sensitivity of 100% and specificity of 92%.
Our results demonstrate excellent correlation between MEG imaging findings with the IAP for language lateralization, and provide new insights into the spatiotemporal dynamics of cortical speech processing.
Pre-operative determination of hemispheric language dominance is an essential part of surgical planning for patients with medically refractory temporal lobe epilepsy and other candidates for brain surgeries that may approach language cortex. From the middle of the twentieth century through the 1990s, the intracarotid amobarbital procedure (IAP), otherwise known as the Wada test, served as the clinical standard for determining language lateralization (Wada, 1949). However, this invasive test has multiple associated risks, with a complication rate estimated at 10.9% (Loddenkemper et al., 2008). There are significant limitations to IAP testing: the window available for testing is limited and flow of amobarbital into the contralateral hemisphere is not infrequent (Simkins-Bullock et al., 2000). IAP protocols vary among centers, making comparative research difficult to carry out, and the worldwide supply of amobarbital has not been reliable. Moreover, a recent study demonstrated that unilateral injection of amobarbital does not entirely eliminate receptive language function in a majority of patients (Hickok et al., 2008). This suggests that processing of even simple language tasks involves some degree of bilateral activation, calling into question the assumptions on which the IAP relies. Indeed, recent surveys of European and worldwide epilepsy centers, as well as experience at an active US epilepsy surgery program, suggests that the IAP is no longer regarded as the only standard for presurgical language lateralization (Baxendale et al., 2008; Haag et al., 2008; Loddenkemper 2008)
For these reasons, neurosurgical centers are now considering non-invasive functional brain imaging techniques such as functional magnetic resonance imaging (fMRI) and magnetoencephalography (MEG), as alternatives to the IAP. Unlike the IAP, a pseudo-lesion technique that determines the hemisphere essential for language function, functional imaging measures brain activation in both receptive and expressive language networks; if temporal resolution is adequate, as with MEG, EEG, and ECoG, language processing networks may be studied in detail (Papanicolaou et. al., 2006). Because imaging tests may confirm regions involved with but not necessarily essential for language function, it remains unclear how best to compare the results of these activation studies with the results of the IAP. The IAP itself may be imperfect as a gold standard language lateralization test, as mentioned above, adding another layer of difficulty when new protocols are investigated to supersede it. Progress on the standardization and adoption of a common non-invasive replacement for the IAP has been inconsistent and has been limited by differences in resources, although motivation to replace the IAP is high (Loddenkemper 2009).
To address this problem, multiple studies have directly compared language lateralization using functional neuroimaging with results from the IAP in epilepsy patients. Several studies have reported the use of fMRI for determining language lateralization with strong correlations with IAP results (Desmond et al., 1995; Binder et al., 1996; Binder et al. 2011; Lehericy et al., 2000; Woermann et al., 2003; Dym et al. 2011; Jones et al. 2011; Liu et al. 2009). However, fMRI lacks the temporal resolution to accurately track the dynamics of language lateralization.
MEG has also been studied as a non-invasive alternative to the IAP. The majority of language MEG studies have determined language lateralization based on dipole analysis (Papanicolaou et al., 2004, 2005; Kamada et al., 2007; Doss et al., 2009). These dipole-based studies have demonstrated 69–86% concordance with the IAP (Papanicolaou et al., 2004; Doss et al., 2009). In response, some epilepsy centers have begun to use MEG with dipole-based analysis in lieu of the IAP for language lateralization in select patients (Doss et al., 2009). However, dipole analysis is a procedure that is well known to be highly sensitive to noise, initialization, and selection of model parameters. The emerging method of sequential single-dipole fitting is inadequate to model activity across multiple brain regions that are expected during complex language tasks, and the neural mechanisms underlying this method are unclear.
Newer algorithms for the analysis of MEG data allow millisecond-by-millisecond estimation of event-related changes in oscillatory activity across multiple frequency bands in three-dimensional brain space (Robinson, 1999; Sekihara et al., 2001; Sekihara et al., 2004), a technique we refer to here as magnetoencephalographic imaging (Dalal 2008). A few studies have recently applied some of these algorithms with enhanced spatial, spectral and frequency resolution to the problem of presurgical language lateralization (Hirata et al., 2004; Kim and Chung, 2008; Hirata et al., 2010). In these studies, task-evoked changes in the MEG power spectrum have identified a network of cortical language regions during language tasks. These changes in power are thought to correlate with cortical activation and are well-described in the EEG (e.g. Pfurtscheller et al. 1993, 1996; Neuper and Pfurtscheller 1996), electrocorticography (ECoG) (e.g. Ohara et al. 2000; Crone et al. 1998a; Crone et al. 1998b; Pfurtscheller et al. 2003; Leuthardt et al. 2004; Szurhaj et al. 2005; Miller et al. 2007), and MEG (Kaiser et al. 2004; Miyanari et al. 2006; Vidal et al. 2006; Gunji et al. 2007; Dalal et al. 2008) literature. While ECoG studies of language event-related power changes have focused on tracking high gamma frequency (> 40 Hz) activity (e.g. Crone et al., 2001; Tanji et al. 2005; Sinai et al., 2005; Canolty et al., 2007; Edwards et al., 2010), MEG imaging studies have demonstrated that cortical language areas in the posterior inferior frontal and middle frontal gyri show event-related power decreases in beta (13–25Hz) (Hirata et al., 2004; Kim and Chung, 2008; Hirata et al., 2010) and low gamma frequency bands (25–50Hz) (Hirata et al., 2004; Hirata et al., 2010), all thought to be a surrogate marker for cortical activation.
However, converging evidence suggests that brain activity during language processing is dynamic, evolving rapidly over multiple regions of known language cortex and dictated by the nature of the language task (Hickok and Poeppel, 2007; Bastiaansen and Hagoort, 2006; Edwards et al., 2010), and this evidence have never clearly been demonstrated in MEG imaging studies of language lateralization. Despite the knowledge gained by human ECoG and MEG studies over the past decade, we are still working to understand the dynamics of spectral power changes that occur bilaterally in cortical regions subserving language. There is no consensus as to which parameters will most accurately reflect hemispheric language dominance. Additionally, no detailed study has used MEG imaging to clearly define discrete time-windows when hemispheric function is maximally differentiated between hemispheres in both expressive and receptive language cortices; to do this, a language task that engages both regions must be used.
The aim of the current study was to accurately characterize the dynamics of language processing using magnetoencephalographic imaging during an auditory verb generation task, a semantic association task that activates both receptive and expressive language networks. Based on this analysis, we then looked to develop a noninvasive, preoperative paradigm to determine language lateralization with excellent predictive value in a heterogeneous population of neurosurgical candidates. The parameters for this paradigm were derived iteratively using a retrospective cohort who had had both IAP and MEG imaging performed, and then tested prospectively in a new cohort. The paradigm was further tested in a cohort of healthy controls who had not undergone IAP testing. Sensitivity and specificity of MEG-imaging was calculated assuming IAP to be the reference standard.
For the development of a clinical protocol for language lateralization determination using MEG, we retrospectively analyzed the MEG data from 21 consecutive subjects who had MEG language testing in the UCSF Biomagnetic Imaging Laboratory (BIL) from 2008–2010. Inclusion criteria: consecutive subjects with medically refractory epilepsy (primarily) or arteriovenous malformations (AVM) who were surgical candidates for removal of the epileptogenic zone or the AVM and who had previously undergone IAP for clinical purposes. The time between IAP and MEG testing varied from a few months to 2 years. Subjects were either having an MEG for clinical purposes and had the language-testing run after the clinical testing was completed, or were separately scheduled for a research MEG during which the language testing was run. Ages ranged from 15 to 56 with an average age of 31 (SD +/− 11.5). Seven were left-handed and 14 were right-handed. Exclusion criteria included 1) if there were severe artifactual activity that could be noted in the MEG sensor array (peak-to-peak fluctuations in spontaneous activity exceeding 2 pico-Tesla) and 2) if subjects were unable to perform the language task during practice. However, no subject met these criteria in our cohort. The IAP was performed by a trained clinical neuropsychologist (DAC-W), based on an established IAP testing protocol (Loring et. al., 1994). IAP results indicated language dominance in the left hemisphere for 13, left greater than right for 2 and right or right/bilateral for 6. For purposes of this study, L>R and right/bilateral were grouped with left and right IAP subjects, respectively. Clinical data for this retrospective cohort are shown in Table 1. Pathological diagnoses are described when known. Many of the subjects were patients referred from the UCSF Department of Neurology for interictal spike mapping as part of a pre-surgical workup.
A second group of subjects with epilepsy and brain mass lesions was included in a prospective study, testing the protocol developed using the retrospective cohort. These were subjects who were awaiting brain surgery for intractable epilepsy or tumor resection, and were scheduled to have IAP for clinical reasons but had not yet done so; as for the other group, some were undergoing MEG for clinical reasons and others were studied for research purposes only. Fourteen consecutive subjects were included, ranging in age from 19 to 52, with an average age of 29 (SD +/− 9.0). Four were left-handed, 7 were right-handed, and 3 were ambidextrous. Within this group, 12 showed left-, one showed right-, and one showed R>L/bilateral-hemispheric dominance for language during the IAP. Clinical data for this prospective cohort are shown in Table 2.
Finally, the language lateralization protocol was tested in a third group of 21 healthy controls without IAP results. Ages ranged from 19 to 45, with an average age of 30; 18 were right-handed and 3 were left-handed.
Informed consent for the study was obtained from all subjects. MEG studies were performed under a protocol approved by the UCSF Committee on Human Research.
Magnetic fields were recorded in a shielded room using a whole-head MEG system (Omega 2000, MEG International Services Ltd. (MISL), Coquitlam, BC, Canada) consisting of 275 axial gradiometers and 29 reference sensors used for computing synthetic third-order gradiometer measurements. The MEG signals were collected continuously and digitized at a sampling rate of 1200 Hz. Head localization was performed at the beginning and ending of the collection to register head position and to measure head movement during the task.
Each individual lay on the MEG bed with his or her head fit snugly inside the MEG dewar helmet. The task, as shown in Figure 1, consisted of 100 nouns presented at a comfortable volume through earphones every 4 seconds. Subjects were instructed to think of a verb or “action word” associated with the noun and to speak into a microphone at the foot of the bed. Auditory stimuli and spoken responses were recorded on separate analog-to-digital channels (ADCs) for post-processing analysis.
Analysis was carried out by and experienced MEG technologist (AMF) and an epilepsy fellow (JBW) under the guidance of SSN, a neuroscientist and bioengineer who directs the BIL, and HEK, an epileptologist and clinical neurophysiologist who is clinical director of the BIL; they were blinded to IAP results during analysis. Onsets of auditory stimuli and verb responses were marked using amplitude threshold detection on the ADCs and verified by eye. Data was formatted into separate trials for each noun-verb pair, excluding trials without a response. Artifact-detection was done by visually examining all trials for many different sensors. Trials with eye blink, EMG artifact, obvious interictal spiking or other obvious artifact were removed. In cases with artifact across most trials, a high pass filter between 1 and 15 Hz was applied before artifact detection in order to preserve enough trials for analysis. Also, trials with speaking within 300 msec prior to or during auditory stimulus presentation were removed.
Evoked MEG responses were examined using adaptive spatial filtering, with analysis time-locked to either on the auditory noun stimulus (“stimulus-locked”) or spoken verbal response (“response-locked”). A multiple spheres head model was calculated using software from MISL. MEG data was bandpass filtered in the beta frequency range (12 to 30 Hz). For the stimulus-locked analysis, sensor covariance was computed using eight 300-msec time windows between 0 and 1000 msec following auditory word stimulus, designated as “stimulus-locked” active periods. For the response-locked analysis, sensor covariance was computed using fourteen 300-msec time windows between 1000 msec prior to and 600 msec following spoken verb response, designated as “response-locked” active periods. The 300-msec time windows overlapped each other by 200 msec; for example, windows for the stimulus-locked analysis were 0 to 300 msec, 100 to 400 msec, etc. For the control period for both conditions, sensor data covariance was computed in a 300-msec window immediately prior to word stimulus. Details of the adaptive spatial filtering algorithm are described elsewhere (Robinson, 1999; Sekihara et al., 2001; Sekihara et al., 2004). In brief, an MEG-based estimate of the source power at each voxel in the brain is computed for each active and control time period using a forward field, which has been computed assuming a multiple local-sphere spherical volume conductor model, making use of the sensor data covariance. Source power estimates are obtained at a 5-mm resolution across the entire brain for the active and control periods, and a pseudo-F ratio is calculated. For these tasks, negative values of the pseudo-F ratio indicate a decrease in beta band power, also referred to in the literature as an event-related desynchronization (ERD); positive values indicate an increase in beta band power, also referred to as an event-related synchronization (ERS) (Hirata et al., 2004; Crone et al., 2001). This process of adaptive spatial filtering was done using a commercially available synthetic aperture magnetometry (SAM) software package (MISL) and replicated and integrated with Nutmeg, a custom-built, in-house software package. Using Nutmeg time-frequency tools, images of beta-band power changes were averaged across time at each voxel and displayed. Stimulus-locked results ranged from 150 to 850 msec following auditory stimulus; response-locked results ranged from 850 msec before to 450 msec after onset of verbal response. Results for each subject and for both conditions were overlaid on MRIs, which were also normalized to standard MNI (Montreal Neurologic Institute) brain space using SPM2 (Tzourio-Mazoyer, 2002) (http://www.fil.ion.co.uk/spm2). Normalization was verified by eye, comparing ventricles and other anatomical boundaries. The transformation matrix derived from this normalization was then applied to each individual subject’s MEG imaging results. Prior studies have described Nutmeg time-frequency analysis in detail (Dalal et al., 2005, 2007, 2008).
A volume-of-interest (VOI) based analysis was developed to capture average power changes over specific brain volumes for each subject. Voxels within each spatially normalized MEG volume were tagged with MNI labels corresponding to anatomical structure. Two large VOIs were created, modeled after the VOIs used in a previous magnetoencephalographic imaging study (Hirata et al., 2004). “VOI-TP” contained voxels labeled as superior temporal gyrus or supramarginal gyrus; “VOI-F” contained voxels labeled as inferior frontal gyrus, middle frontal gyrus or pre-central gyrus. Images were masked with each VOI and an average of all power changes in each voxel was computed for each time and frequency window. Figure 2 shows left and right VOI-TP (A) and VOI-F (B) for a single patient with left hemispheric dominance for language by IAP. Figure 2A shows power changes in the beta band (12–30 Hz) at 650 msec following auditory stimulus in the stimulus-locked condition. Figure 2B shows power changes for the same frequency band at 850 msec prior to speech onset in the response-locked condition. Average F-values, proportional to net power change, for each VOI and time-frequency window and for both stimulus- and response-locked conditions were calculated for all subjects. Subjects were separated into two groups depending on IAP results. Patients with L>R language lateralization for IAP were grouped with left IAP patients for this step and right/bilateral patients were grouped with right IAP patients. All IAPs were performed by an experienced neuropsychologist (DC-W) using the Medical College of Georgia protocol (Lee et al 2002); she was blind to the results of the MEG imaging at the time of IAP for the prospective cohort.
A laterality index (LI) was calculated for each time-frequency window for both analysis conditions using the average F-value of each VOI. This LI helps address the problem of excessive false positives in functional imaging studies for language lateralization, given the extent of bilateral activation that is commonly seen (e.g. Papanicolaou et al., 2004). Under the assumption that a power decrease in the beta band is a marker for cortical activation, and that greater beta-power decrease in one hemisphere versus the other is a marker for stronger lateralization of function, we used the following formula: LI = −1 * (L − R)/(|L| + |R|), where L represents the averaged F-value in the left VOI and R represents the averaged F-value in the right VOI. An LI value of +1 or −1 would indicate greater beta-power decrease in the left or right hemisphere, respectively. The LI was calculated for both VOIs across time and frequency and then averaged for each group of patients (i.e., right and left IAP). Average LI values were plotted as a function of time for each frequency band, and time windows with significant differences between right and left IAP groups were determined using the retrospective cohort.
A “stimulus-locked” LI was calculated using the three most significant time points for the stimulus-locked condition within VOI-TP, and a “response-locked” LI was calculated similarly for the response-locked condition within VOI-F. An overall, “combined” LI was also calculated by averaging the stimulus- and response-locked LIs, and compared with each patient’s IAP result.
Sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were determined. For this, IAP results were grouped into two categories: 1) right or R>L (“right IAP”) and 2) left or L>R (“left IAP”). MEG results were classified using an empirically-determined, combined LI threshold of ±0.1, i.e. MEG was classified as right for LI < −0.1 or left for LI > 0.1. MEG results falling between −0.1 and 0.1 were considered to be either L>R or R>L; for example, a right IAP would be in agreement with an MEG combined LI of 0.05, since LI of 0.05 could be either R>L or L>R. Further assumptions are as follows: right IAP/right MEG is a true positive (TP); left IAP/left MEG is a true negative (TN); right IAP/left MEG is a false negative (FN); and, left IAP/right MEG is a false positive (FP). Using these definitions, sensitivity is calculated as TP/(TP + FN), specificity is calculated as TN/(FP + TN), PPV is calculated as TP/(TP + FP), and NPV is calculated as TN/(FN + TN).
The MEG language lateralization protocol established using the retrospective cohort was also applied to the prospective cohort (Table 2) and to the group of healthy controls. Results for patients in the prospective cohort were compared to their IAP language results.
Subjects tolerated the MEG language testing well; there were no adverse events. There were no complications to IAP in the cohorts.
Three-dimensional overlays of average beta-band (12–30 Hz) power changes in the stimulus-locked and response-locked conditions are shown for the retrospective group in Figures 3 and and4,4, respectively. Individual beta-power changes were spatially normalized to MNI space and averaged within each group for each condition. The group-averaged beta-power changes were then thresholded at half of the absolute maximum F-value over the shown time course and displayed on 3-D rendered brains in MNI space. Deep sources are not projected to the surface in this rendering. Figures 3A and and4A4A show averages for 15 subjects with left hemispheric language by IAP (“left IAP subjects”) and Figures 3B and and4B4B show averages for 6 subjects with right hemispheric language by IAP (“right IAP subjects”) in the retrospective cohort.
Following onset of auditory word stimulation (hearing a noun), the average pattern of event-related power change in this cohort of left IAP subjects showed beta-power decrease beginning predominantly in the left inferior frontal region 250 to 350 msec after stimulus (Figure 3A). This was followed by beta power decrease in bilateral, LR peri-sylvian and superior temporal regions including Wernicke’s area. As this resolves, there is next an enhanced beta-power decrease in the left inferior frontal area 850 msec after stimulation. Right IAP subjects also show beta power decrease in the inferior frontal region (Figure 3B), but with early bilateral involvement (right greater than left) 250 msec following stimulation. An initial beta-power decrease in the right superior temporal region is seen at 350 msec; this region continues to broaden and remains bilateral (right greater than left) from 350 to 750 msec.
In left IAP subjects, verb generation resulted in a highly lateralized beta-power decrease in the left inferior and middle frontal regions (Figure 4A). This was restricted primarily to the left hemisphere up until 250 msec prior to vocalization, when bilateral beta-power decrease in the precentral gyrus was observed. Right IAP subjects demonstrated bilateral beta-power decrease throughout this recording period (Figure 4B), with stronger beta-power decrease in the right inferior and middle frontal regions.
Time courses of the average F-values within each VOI are shown for the retrospective cohort in Figure 5. Figures 5A and 5B show VOI-TP (STG plus supramarginal gyrus) and VOI-F (IFG, MFG, pre-central gyrus), respectively, for the stimulus-locked condition; Figures 5C and 5D show VOI-TP and VOI-F, respectively, for the response-locked condition. Blue and pink lines correspond to average F-values for left and right IAP subject groups, respectively. Solid and dashed lines correspond to averages in the left and right hemisphere VOIs, respectively. An asterisk next to a point indicates statistical significance using a 2-tailed t-test between left and right IAP subject groups (p<0.05). For the stimulus-locked condition, significant differences between right hemisphere beta-power decrease in right and left IAP groups occurred at time points 750 and 850 ms for VOI-TP (5A) and at 850 ms for VOI-F (5B). For the response-locked condition, significant differences between right hemisphere beta-power decrease in right and left IAP groups occurred at time points −850 through +150 ms for VOI-TP (5C) and −850 through −150 ms for VOI-F (5D). For the left hemisphere, no differences were seen between subject groups in any condition.
Figure 6 shows time courses of the mean LI for the stimulus-locked (6A and 6B) and response-locked (6C and 6D) conditions for both VOIs. Lighter lines above and below the mean indicate standard error. Blue and pink correspond to left and right IAP patient groups. Asterisks indicate time points that are statistically different between left and right IAP patient groups with p<0.001. For the stimulus-locked analysis, LIs of left and right IAP groups were significantly different with p<0.001 at 650 through 850 msec within VOI-TP, and at 850 msec within VOI-F. Additional time points demonstrated significant differences between groups with p<0.05 at 450 msec within VOI-TP and at 450, 650 and 750 msec within VOI-F (Figures 6A and 6B). For the response-locked analysis, LIs were significantly different with p<0.001 at −850 and −650 ms within VOI-F only. Additional time points were significant with p<0.05 at −850 through −50 ms for VOI-TP and at −750, and −550 through −250 ms within VOI-F (Figures 6C and 6D).
For each condition, a cluster of time points showing greatest significance between IAP groups was selected to use as an overall index of lateralization. For the stimulus-locked condition, this occurred at time points 650, 750 and 850 msec in VOI-TP (STG, supramarginal gyrus) following auditory stimulus. LIs for these time points were averaged to generate the “stimulus-locked LI” for each subject. For the response-locked condition this cluster occurred at 850, 750 and 650 msec prior to verb generation. LIs for these time points were averaged to generate the “response-locked LI” for each patient. In both VOI-TP and VOI-F for both stimulus- and response-locked conditions, there were other points showing significance with p<0.05; however, these two clusters included the time points most strongly distinguishing left and right IAP patients.
Figure 7 shows the distribution of stimulus-locked (first column) and response-locked (second column) LIs for each patient. The combined LI measure, an average of the stimulus-locked and response-locked LIs, is shown in the third column. Blue and pink boxes in Figures 7A and 7B correspond to left and right IAP patients, respectively. Figure 7A shows results for the retrospective group, 7B shows results for the prospective group, and 7C shows results for the control group. In Figure 7C, green boxes indicate right-handed subjects while purple boxes indicate left-handed subjects. Dotted lines connect the values obtained for each patient. For LI, a value of +1 indicates strong left-hemispheric lateralization, while −1 indicates strong right-hemispheric lateralization.
In the retrospective cohort (Figure 7A), using an LI threshold of ±0.1, MEG shows bilateral (i.e., L>R or R>L) lateralization for −0.1 ≤ LI ≤ +0.1, left lateralization for LI > +0.1 and right lateralization for LI < −0.1. Thirteen out of the 15 left IAP subjects showed left lateralization for the stimulus-locked, response-locked and combined LIs. For the right IAP group, two subjects, including one “right/bilateral” subject, showed right MEG lateralization for all 3 LIs. Three other subjects showed bilateral activity for either the stimulus-locked or response-locked LI with a combined LI showing right MEG lateralization. The sixth subject showed bilateral stimulus-locked LI, left response-locked LI, and bilateral lateralization for the combined LI measure (see Table 3). Of note, two members of the retrospective cohort had combined LIs between −0.1 and +0.1, classifying them as bilateral (either L>R or R>L). One of these subjects was classified as “L>R, bilateral” with IAP, and the other as having right-sided hemispheric dominance.
In the prospective cohort (Figure 7B), using the same thresholds, 11 of 12 left IAP subjects showed left and one showed right lateralization for stimulus-locked, response-locked and combined LIs. The one subject with “right/bilateral” hemispheric lateralization on IAP showed right-lateralization for the stimulus-locked LI, left-lateralization for the response-locked LI, and right lateralization for the combined LI (LI = −0.11). The subject who was right dominant by IAP had all three MEG LIs suggesting right sided lateralization. See Table 4 for a list of LI values and IAP results for each patient.
In a control group (Figure 7C), again using the same thresholds, 16 of 21 subjects showed left lateralization for all LIs. Three showed bilateral stimulus-locked LI, with left response-locked and combined LIs and one showed right response-locked LI with left stimulus-locked and combined LIs. One subject who was left-handed showed right LIs for all 3 measures. The other two left-handed subjects showed left lateralization for all 3 LIs. See Table 5 for a list of LI values for each control subject.
In the retrospective cohort, sensitivity, specificity, PPV and NPV were all 100%. For the prospective cohort, sensitivity was 100%, specificity was 92%, PPV was 67% and NPV was 100%. Combining the retrospective and prospective groups together (35 patients, including 8 with right or R>L IAP), sensitivity and NPV were 100%, specificity was 96%, and the PPV was 89%.
Results for selected individual subjects are described below. Several subjects were noted to have interhemispheric dissociation of lateralization during receptive (stimulus-locked) and expressive (response-locked) tasks. In addition, one subject in the prospective group had a mismatch between lateralization with the IAP and with MEG imaging (see Tables 3 & 4).
This is a 44 year-old, right-handed woman with medically refractory, focal seizures originating in the right temporal lobe. IAP testing demonstrated L>R hemispheric language function. During MEG imaging, the patient demonstrated beta-power decrease maximal on the right during the stimulus-locked period (LI −0.73) within receptive language areas (VOI 1). However, the patient was found to have beta-power decrease maximal on the left during the response-locked period (0.84) within expressive language areas (VOI 2). The combined LI was 0.06, a value consistent with bilateral beta-power decrease with a slight left predominance. The patient had a right temporal resection, and was found to have a right temporal cortical dysplasia. No post-surgery follow-up data was available.
This is a 30 year-old, ambidextrous man (writes with right hand, throws with left) with localization-related epilepsy secondary to a left mesial temporal mass. Motor testing during MEG imaging demonstrated event-related beta-power decrease in bilateral motor cortex with right index finger movements and beta-power decrease in contralateral motor cortex with movement of the left index finger (see Nagarajan et al. 2008 for method). On IAP testing the patient was found to have bilateral (R>L) language function. During MEG imaging, the patient demonstrated beta-power decrease maximal in right hemisphere during the “stimulus-locked” period (−0.32) and maximal in the left hemisphere during the “response-locked” period (0.11). Combined LI result was −0.11, a value consistent with right lateralized beta-power decrease. The patient had a resection of the left temporal mass; pathology was suggestive of a low-grade glioma. There was no post-operative language deficit on clinical examination.
A 38 year-old, ambidextrous man with localization-related epilepsy secondary to a left mesial temporal mass extending into the subinsular region. IAP testing demonstrated left hemisphere dominant language. During MEG imaging, the patient demonstrated right-side predominant beta-power decrease during both the stimulus-locked (−0.47) and response-locked (−0.65) periods, resulting in a strongly right lateralized combined LI of −0.56. Pre-operative auditory evoked fields showed primary auditory activation in the left hemisphere posterior to the lesion. The patient had a resection of the left mesial temporal mass; pathology showed a gemistocytic astrocytoma. There was no post-operative language deficit on clinical examination.
Our findings demonstrate that hemispheric dominance of language is a dynamic process that can be reliably observed with magnetoencephalographic imaging during an auditory verb generation task. We found significant beta-power decreases in both the dominant and contralateral cerebral hemispheres that were bilateral for the first few hundred milliseconds following auditory speech presentation, and hundreds of milliseconds prior to overt speech production. Consistent and significant lateralization is only observed during a window of time between speech perception and production. A detailed analysis of this process, with iterative extraction of parameters that best discriminated left and right IAP subject groups, resulted in the development of a highly sensitive and specific approach to language lateralization using MEG imaging. The reliability of this method was subsequently assessed prospectively in a heterogeneous population of neurosurgical patients and we found excellent correlation with results from the IAP in this independent cohort.
In the present study, subjects with left-hemisphere dominance by IAP demonstrated beta-power decrease in the left inferior and middle frontal regions prior to verb generation. This was seen up until 250 msec prior to vocalization, when beta-power decrease increased bilaterally and spread into the precentral regions. Patients with right-hemisphere dominance by IAP demonstrated bilateral beta-power decrease in the inferior and middle frontal regions during the same time period. Language lateralization was highly correlated with IAP results during both word-stimulation and verb-generation periods; maximal correlation was seen at 650–850 msec following word stimulation and 650 and 850 ms prior to vocalization. These findings are consistent with language models that describe expressive language as a highly lateralized process (Hickok and Poeppel, 2004), and are similar to fMRI studies that have demonstrated maximum language lateralization in the inferior frontal gyrus (Bizzi et al., 2008; Liljestrom et al., 2009; Dym et al. 2011, Binder et al. 2011; Liu et al. 2009), as well as another MEG study using current density imaging (MR-FOCUSS) that reported consistent lateralization in Broca’s area during a picture naming task (Bowyer et al., 2005).
A recent study used ECoG to assess the dynamics of cortical activation in right-handed epilepsy patients performing a verb generation task (Edwards et al., 2010). Findings within individual subjects demonstrated power changes in beta and gamma frequency bands that were similar spatially and temporally to the activity distributions seen in the current MEG study (Figs 3–4). In both studies, power changes were seen in the peri-sylvian cortex (STG, supramarginal gyrus) 300 msec following auditory word stimulation, presumably representing the previously described “ventral stream” of acoustic-to-semantic speech processing (Hickok and Poeppel, 2007). This activity was followed by power changes in the inferior frontal gyrus 700 msec after stimulation, and in the inferior frontal and peri-rolandic regions 100–300 msec prior to verb generation (Edwards et al., 2010). Regions with later power changes are presumably components of the “dorsal stream” of speech processing, implicated in phonological encoding and preprocessing for articulation (Hickok and Poeppel, 2007), consistent with patterns observed in our study.
Some earlier MEG language lateralization studies using dipole analysis have demonstrated good correlation with the IAP based on enumerating dipoles within the posterior superior temporal region (Papanicolaou et al., 2004). MEG signal in this region has been shown to localize to regions of electrocorticographic (ECoG) activity during language tasks (Castillo et al., 2001). However, the validity of using sequential single dipoles to represent complex cortical activation patterns associated with language tasks, and difficulty in inferring neural mechanisms underlying these methods, motivate exploration of other techniques for MEG imaging of language. Other recent language lateralization studies have used spatial-filtering algorithms with MEG and have demonstrated strongly lateralized task-evoked decreases in beta- and low-gamma power in the inferior frontal region (Hirata et al., 2004; Kim and Chung, 2008; Hirata et al., 2010; Pang et al. 2011). While data from these prior studies have not clearly shown that hemispheric dominance may be a dynamic process, the present study confirms and elaborates upon the findings of robust beta-power decrease across a wide fronto-temporal network of brain regions during an expressive language task and the bihemispheric dynamics of these oscillations. The current study also extends recent work by Hirata et al. (Hirata et al. 2010) who showed concordance with IAP findings in 86% of subjects using a single time-window of analysis in the receptive language period of 0–1000ms following word-onset in a covert speaking task. In contrast to their study, where the words were presented visually, we focus here on auditory presentation of words. Because their study used a word-reading task, their results may reflect orthographic encoding rather than phonological and linguistic processes. In contrast to their report, our study shows that posterior superior temporal regions indeed also exhibit language lateralization when examined at appropriate time windows. Although Hirata et al. report multiple event-related frequency band power changes, they specifically observe lateralized power changes in the beta-band, as in our study.
Further analysis of beta-power decrease averaged across subjects demonstrated that the difference in average F-values for left- and right-IAP subjects was seen only in the right hemisphere. This finding suggests that determination of language laterality using MEG imaging is driven primarily by the degree of beta-power decrease seen in an individual’s right hemisphere. It remains unclear what features of right hemisphere beta-power change are sufficient and necessary to result in lateralization (e.g., degree of lateralization, intensity of beta-power decrease, spatial distribution of beta-power decrease). Recent evidence suggests that the right hemisphere is critical to language functioning, even in right-handed subjects with traditional “left dominant” language (van Ettinger-Veestra et al., 2010). These observations further suggest that a purely hemispheric approach to language lateralization, e.g. the IAP, is inadequate to characterize basic language networks that frequently utilize the contralateral or “non-dominant” hemisphere.
In our study, an approach to language lateralization was developed based on analysis of the maximally sensitive time-windows within cortical regions-of-interest for both receptive and expressive language. Language lateralization during the stimulus time-locked recording window (receptive language) for the retrospective cohort demonstrated a test sensitivity of 100% and specificity of 93% when compared to results of the IAP. During the verb generation time-locked recording window (expressive language), sensitivity and specificity were 83% and 93% respectively. These values are comparable to or somewhat improved from those found in prior studies utilizing MEG for language lateralization (Hirata et al., 2004; Papanicolaou et al., 2004; Kim and Chung, 2008; Doss et al., 2009; Hirata et al., 2010).
Of note, combined analysis of beta-frequency power change in receptive and expressive language areas demonstrated correlation with IAP language lateralization results in 34 of 35 (97%) patients (combining retrospective and prospective cohorts). A sensitivity and specificity of 100% and 96%, respectively, were calculated using a laterality index (LI) threshold of 0.1. These findings suggest that an inclusive approach to language analysis, combining analyses of receptive and expressive language function, more closely matches language lateralization by IAP. Although ours was a small population, the sensitivity and specificity may be comparable to that seen in fMRI studies of language lateralization (Desmond et al., 1995; Binder et al., 1996; Woermann et al., 2003; Arora et al., 2009).
To further validate our findings, a second group of patients was studied using the objective measures developed in the first group. Thirteen of fourteen subjects (including two right-side language dominant subjects) demonstrated language lateralization corresponding to their IAP result. This heterogeneous group of subjects included those with large tumors, which are known to distort the anatomical representation of language (Wellmer et al., 2009) and which can also interfere with fMRI signal, particularly in lesions with aberrant vasculature (Holodny et al., 1999; Lehericy et al., 2002; Grummich et al., 2006, Wellmer et al., 2009). These findings again achieved excellent correlation with the IAP by focusing on time periods in which the dynamic process of language processing is maximally lateralized.
An additional group of 21 neurologically normal subjects were tested using our paradigm. All 18 right-handed and 2 left-handed subjects demonstrated left-side language dominance; the one case of right-side language dominance was seen in a subject who was left-handed. Although MEG language dominance could not be compared to IAP for these healthy controls, these findings conform to the expected distribution of language dominance within left and right-handed subject groups (Victor, 2001). This finding supports the idea that our MEG language paradigm may generalize widely to other clinical populations. Given that prior studies have demonstrated significant alteration of the topographic representation of language in people with epilepsy and mass lesions (Pataraia et al., 2005; Breier et al., 2005; Wellmer et al., 2009), it is important that this method of language lateralization be validated in a wide array of subjects.
Overall, maximal lateralization of beta-power decrease was seen both in temporal and in inferior and middle frontal regions, a finding in agreement with several fMRI studies of language lateralization during expressive language tasks (Kamada et al., 2006; Salmelin et al., 2007; Bizzi et al., 2008), as well as during rest (Liu et al. 2009). Functional MRI studies performed with receptive language tasks have less reliably lateralized language (Salmelin et al., 2007). These findings suggest that both fMRI and MEG can reliably lateralize language in the inferior frontal region. At present, routinely available fMRI research scanners are unable to evaluate event-related signal changes occurring under time periods of 1 second (Babiloni et al., 2009). It is therefore important that MEG protocols for language lateralization are designed differently from fMRI studies, taking advantage of discrete time windows that accurately reflect the dynamic, bilateral process of cortical language processing. Although fMRI lacks the temporal sensitivity to record brief periods of lateralized receptive language activity within the superior temporal region, the combination of fMRI and MEG may present the best tools for estimating hemispheric dominance for language, and future studies combining these modalities are warranted.
This is especially critical in patients with significant brain pathology, as there can be dissociations in laterality between receptive and expressive language function. This has been reported in a patient who showed expressive language function in the left hemisphere with fMRI and receptive language function in the right hemisphere with MEG (Kamada et al., 2006). Additional cases demonstrating a similar dissociation have been reported with fMRI alone (Baciu et al., 2003, Ries et al., 2004). In the current study, four patients demonstrated dissociation of lateralization between hemispheres for receptive (“stimulus-locked”) and expressive (“response-locked”) recording windows (Supplementary document 1). One of these patients was ambidextrous and another was left handed. All four patients had structural lesions (tumor or cortical dysplasia), but no additional unifying features. In all cases, averaging of LIs from “stimulus-locked” and “response-locked” periods resulted in language lateralization corresponding to the patient’s IAP result. We describe, in several subjects, an interhemispheric dissociation between receptive and language function using MEG These cases demonstrate the importance of testing both receptive and expressive components of language with MEG imaging, and analyzing cortical regions responsible for each. By using results from each functional region, a closer approximation is made to the IAP, which analyzes many facets of language (Loring et al., 1994). Had these patients only performed isolated receptive or expressive language tasks, false lateralization may have occurred.
There are several limitations to the current study that need to be acknowledged. Although the total number of patients (35) analyzed is larger than that found in early, comparable studies (Hirata et al., 2004, Kim and Chung., 2008; Doss et al., 2009), only 14 of these patients were studied prospectively. Future studies will need to be performed on larger groups of patients in a prospective fashion. Given the immediate aims of our study, we have yet to correlate our findings with long-term functional language outcomes in the post-operative period. To date, this has been a limitation of all MEG studies of language lateralization (McDonald, 2008).
In addition, the current study used VOIs consisting of broad cortical regions that grossly correspond to receptive and expressive language areas. Future studies may benefit from designation of specific sub-regions that more accurately encompass known language processing centers (Hickok and Poeppel, 2007). By using higher spatial specificity along with discrete time windows, higher test sensitivity may be achieved. This approach would also allow for a more direct comparison between spatial distributions of task-evoked power change using MEG and ECoG recorded during language tasks (Edwards et al., 2010; Hirata et al., 2010).
Our technique for language lateralization using MEG imaging is promising as a replacement for the IAP. This approach may benefit from the concomitant use of other functional imaging modalities such as fMRI. Previous studies have demonstrated increased sensitivity for language lateralization when combining fMRI with MEG dipole analysis or magnetoencephalographic imaging (Grummich et al., 2006; Kamada et al., 2007). A combined approach may be of particular benefit in patients with large vascular lesions, where significant distortion of the BOLD fMRI signal can occur (Grummich et al., 2006; Wellmer et al., 2009). Ideally, this would lead to an algorithmic protocol that offers a tailored combination of non-invasive functional imaging modalities to each neurosurgical candidate depending on lesion type, presumed location, and expected surgical plan, subsequently making the IAP necessary in only the most ambiguous of cases.
The authors thank Maria Ventura for excellent technical assistance and data collection. This work was funded in part from grants from the following grants: NIH R01DC004855, R01DC006435, R21NS076171, NIH/NCRR UCSF-CTSI grant UL1 RR024131 to SSN, NSF BCS0926196 to SSN and JFH, UCSF CTSI/REAC grant to HK and SSN, and a Epilepsy Foundation Fellowship to JA.