- Top of page
- RESULTS FOR CONTROL GROUP
- RESULTS FOR PATIENTS
Summary: Purpose: Functional magnetic resonance imaging (MRI) using two language-comprehension tasks was evaluated to determine its ability to lateralize language processing and identify regions that must be spared in surgery.
Methods: Two parallel cognitive language tasks, one using auditory input and the other visual input, were tested in a group of control subjects and in temporal lobe epilepsy patients who were candidates for surgical intervention. The patient studies provide an opportunity to compare functional MRI language localization with that obtained using Wada testing and electrocorticography. All of the patients in this study underwent all three procedures and a battery of neuropsychological testing. Such studies provide an opportunity not only to validate the fMRI findings but also, by comparing the patient results with those obtained in control subjects, to provide insight into the impact of a pathology such as epilepsy on cortical organization or functional patterns of activation.
Results: The results reveal both modality-dependent and modality-independent language-processing patterns for visual versus auditory task presentation. The visual language task activated distinct sites in Broca's area, BA (Brodmann area) 44 that were not activated in the auditory language task. The auditory language task strongly activated contralateral right BA22-21 area (homologous to Wernicke's area on the left). Language lateralization scores were significantly stronger for visual than for auditory task presentation. The conjunction of activation from the two different input modalities (modality-independent areas) likely highlights regions that perform more abstract computations (e.g., syntactic or pragmatic processing) in language processing. Modality-specific areas (e.g., right Wernicke, left fusiform gyrus, Broca BA44, supramarginal gyrus), appear to cope with the computations relevant to making contact with these more abstract dimensions. Patients showed recruitment of contralateral homologous language areas (p < 0.005) that was significantly above that found in a normal control group. Extra- and intraoperative cortical stimulations were concordant with the fMRI data in eight of 10 cases. The fMRI lateralization scores were also consistent with the Wada testing in 8/10 patients.
Conclusions: The fMRI results demonstrate that the epileptic brain may be a progressive model for cortical plasticity.
Functional magnetic resonance imaging (fMRI) has become an important tool in basic neuroscience, allowing investigation of a wide range of cognitive functions in normal subjects. However, the potential clinical applications of this technology have yet to be fully explored, particularly in neurosurgery. One area of potential value is in using fMRI to obtain information about language lateralization and localization that is currently obtained by using more invasive procedures such as the Wada test and direct electrical stimulation of cerebral cortex. To gain widespread clinical use, it is important to demonstrate that fMRI can provide useful information with the same reliability and accuracy as the more established methods. Functional MRI may allow more extensive mapping of complex language systems than is possible using the Wada test or cortical stimulation techniques and therefore provide the opportunity to understand these systems better. For example, syntactic and pragmatic processing in sentence reading or listening may be at risk in patients who are undergoing surgery in the language-dominant hemisphere, yet the anatomic organization of these processes remains uncertain (1–3). The ability to develop a detailed map of language systems with a noninvasive technique may allow better prediction of clinical outcome after surgery in the language-dominant hemisphere.
Several studies of fMRI language activation have reported left hemisphere lateralization for language in control subjects and patients. However, only a very limited number of studies have validated their fMRI results with Wada testing in the same sample. Springer et al. (4) reported the largest sample of fMRI language lateralization to date. In this study, 100 normal controls and 50 patients with epilepsy were studied using a semantic decision task with auditory presentation of verbal stimuli. However, although they report that the epilepsy patients underwent the IAP (Wada test ), comparison of the laterality as determined by these two techniques is not reported. Nonetheless, they report fMRI-determined laterality in normal subjects that is consistent with reported frequency of left and right hemisphere language dominance in other studies. In an earlier study, Desmond et al. (6) reported fMRI and Wada results in seven patients (3 right hemisphere dominant). Their fMRI task used visually presented words, and required a perceptual or semantic judgment about the subjects. The resulting activation of the inferior frontal gyrus and neighboring cortex (BA 45, 46 and 47) during fMRI was concordant with the Wada result in all 7 cases. Binder et al. (7) also reported a strong correlation between Wada test results and fMRI. In their sample of 22 consecutive epilepsy patients, the correlation between Wada results and a fMRI single-word activation task also suggested that language lateralization was a continuous, rather than a dichotomous, variable.
Determination of language lateralization is an important goal of both the Wada and fMRI. However, more-specific functional maps are often needed to assist with defining the boundaries of language for minimizing postoperative language morbidity. This is accomplished by using direct cortical stimulation of the cerebral cortex during surgery on an awake patient, or with implantation of subdural grids. To increase our understanding of fMRI activation patterns, direct clinical correlation of stimulation-derived functional maps with fMRI language maps is needed. Roux et al. (8) reported a series of 22 patients that underwent cortical stimulation and fMRI. Cortical stimulation for motor function (16 of the 22 patients) was consistent with the fMRI data. However, the remaining six patients had temporal lobe tumors and underwent stimulation for language mapping, for which concordance was poor and limited to anterior (precentral) language areas. Similar studies have been performed comparing positron emission tomography (PET) activation maps with cortical stimulation data (9,10) wherein increased blood flow was observed in PET imaging during both visual and auditory naming tasks, and these regions of increased flow corresponded to regions in which subdural electrodes disrupted language during electrical stimulation.
In general, fMRI language studies have shown promising results, but several issues limit interpretation and consequently the clinical applicability at present. First, the clinical correlations with the Wada test are good, but not perfect. This may be a methodologic issue with respect to the types of language tasks that are used, with fMRI more readily classifying language laterality along a continuum. Second, most of the studies that report Wada and fMRI results are small samples and use various methods, and consequently need replication. In addition, although it is clear that multiple methods can produce activation, the studies have been limited to single-method tasks and may not produce activation in a distribution language system. For example, studies have shown critical areas for word generation, semantic functions, and lexical processing, but only a limited number have established a pattern of whole-sentence processing, which requires both semantic and syntactic processing (11–17). The final difference between functional MRI mapping and Wada testing concerns the issue of which hemisphere is critical to a task (which is defined by Wada) versus regions that are involved in some aspect of a task (which are defined by fMRI, and which may or may not be critical). Comparing fMRI activation patterns with cortical stimulation and Wada results will improve our understanding of the patterns of activation observed in fMRI.
The present study was undertaken with several goals in mind. First, we examined cortical localization for sentence processing in both nonimpaired controls (n = 10) and a sample of epileptic patients (n = 10). The language tasks used were designed to engage broadly those neural circuits necessary for the lexical access, syntactic analysis, and comprehension of both spoken and printed language materials, thereby providing activation of a distributed language system. By examining activations coincident with both the auditory and visual input pathways, we separate perception effects from more direct language-processing effects. However, modality-specific areas are not less clinically important, as damage to such areas also can lead to cognitive deficits. Studies comparing visual and auditory input in language tasks have been demonstrated to be effective in PET imaging (9,10).
The group comparison (controls vs. patients) asks how patients with a history of neurologic impairment differ from controls when performing the same relatively simple language tasks. Most important, our goal is to validate the fMRI language findings in the patient sample by using established methods of determining language lateralization (Wada test) and intrahemispheric localization (cortical stimulation). We contrast both inferred lateralization and within-hemisphere localization from fMRI measures with standard measures including Wada testing and intra/extraoperative cortical stimulation.
RESULTS FOR CONTROL GROUP
- Top of page
- RESULTS FOR CONTROL GROUP
- RESULTS FOR PATIENTS
Consistent areas of activation were observed across subjects during both tasks. Figure 1 illustrates the average activity pattern for all the eight right-handed controls to a depth of 10 mm from the brain atlas surface using normal fusion software (28). Figure 2 shows a representative example of fMRI language mapping for a control subject (patient 6). fMRI lateralization score results on the eight RH healthy volunteers (displayed for t threshold = 1.5 in Fig. 1) showed that the visual language task provided a statistically significant higher lateralization score than the audio language task whether the whole brain or limited ROI laterality approach is used (Table 2). Total laterality scores between right-handed and left-handed normals showed a trend toward greater laterality in right-handed control subjects. Talairach coordinates were calculated for the center of each important region of activation (Table 3)(23).
Figure 1. 3D views of average language activations for the 8 right handed control subjects, superimposed on the Montreal Brain Atlas standard 3D volume. Note the increased activation in the left and right STG for the auditory task relative to the visual task, and the increase in the left fusiform gyrus, SMA, preCentral sulcus and BA 44 in the visual relative to the auditory task.
Download figure to PowerPoint
Figure 2. Illustration of typical language areas activated by both tasks as well as those areas activated only in the visual or auditory task. It shows two examples of 2D coronal oblique (−15°) activation maps: for the control subject #6 (column 1, 2, 3) and for the patient subject #17 (column 4, 5, 6). 1st and 4th columns correspond to the Visually presented language, 2nd and 5th columns to the Auditory presented language, and 3rd and 6th to the common area between the two (logical AND operation). For the normal subject, note the activations in the left fusiform gyrus, BA 44, SMG, and preCentral sulcus for the visual language task (1st column), in the right STG for the auditory task (2nd column), and in the common areas for the logically ANDed analysis (3rd column). For the left temporo-parietal epileptic patient, note the bilateral activations in both tasks (4th and 5th column).
Download figure to PowerPoint
Table 2. Subject laterality scores and Wada results
|Subj. No.||Handedness||Wada||Whole brain laterality||BA44/45 and post-BA22 laterality|
|Visual||Auditory||Vis. & aud.||Visual||Auditory||Vis. & aud.|
|RH controls (mean ± std)||74 ± 11||44 ± 22||87 ± 18||82 ± 25||58 ± 27||87 ± 18|
|RH patients (mean ± SD)||41 ± 35||2 ± 35||58 ± 39||39 ± 29||36 ± 46||62 ± 21|
Table 3. Talairach coordinates (x,y,z) of centers of activated regions
|Supramarg gyrus BA 40||−48, −46, +35||Poor activation||Poor activation|
|Wernicke's area BA 22||−52, −8.5, +8.5||±55, −7, +5||−59, −13, +9|
|Fusiform gyrus||−42, −27, −20||Poor activation||Poor activation|
|Inferior frontal G BA 45||±49, +22.5, +12.5||±46, +25, +14||−51, +21, +12|
|Inferior frontal G BA 44||±49.5, +14, +22.5||±52, +13, +25||−54, +13, +23|
|Precentral sulcus||−46.5, −2, −42||Poor activation||Poor activation|
|Supplementary motor area||−5, +7, +58||Poor activation||Poor activation|
Cortical activations of the left inferior frontal gyrus (IFG) in its opercular part (Broca's area) were present in all subjects for both the VLTask and the ALTask. Activations in BA44 were more frequent and stronger on the left for the VLTask (n = 9/10) than for the ALTask (n = 2/10) whereas both tasks activated BA45 (n = 10/10). Quantitative comparison showed significant differences, p < 0.05 (Table 4) in BA 44. The left precentral sulcus (PrS) was activated in all subjects for the VLTask and in half of the subjects for the ALTask. Left supplementary motor area (SMA) was activated mostly in the VLTask in eight of 10 patients versus three of 10 for the ALTask.
Table 4. Quantitative assessments
|Region of interest||Quantification of Activation||Surface of the ROI (cm2)|
|Left BA 22||649.5||830.9||>0.05||7.8|
|Right BA 22||125.4||615||<0.01||6.2|
|Left BA 45||673.4||607.4||>0.05||1.8|
|Left BA 44||1,164.7||587.2||<0.05||2.2|
A robust activation pattern emerged in the left BA22 [Wernicke's area; superior temporal gyrus (STG)] in all subjects for both tasks (n = 10/10). Activation was observed in the higher-order auditory cortex in the auditory task, as would be expected, including BA42 (superior temporal planum) or BA41 (Heschel's gyrus/primary auditory cortex) on either side during the ALTask, and the auditory task also preferentially activated posterior BA22 bilaterally, partly contributing to the more bilateral scores found with the auditory language task. Additional right posterior BA22 activation was far more frequent and stronger for the ALTask (n = 6/10) than for the VLTask (n = 0/10). Quantitative comparison showed significant differences between these tasks in this region (Table 4). The left middle temporal gyrus (MTG, BA21), was activated by the VLTask, whereas bilateral BA21 activations were present in all volunteers for the ALTask. Left fusiform gyrus (FuG, T4) activation was present in 70% of the subjects and only for the VLTask. The left supramarginal gyrus (SMG, BA40) was activated only in the VLTask. The angulate gyrus, known to play a role in reading, is not included in the scanning range of this study.
- Top of page
- RESULTS FOR CONTROL GROUP
- RESULTS FOR PATIENTS
In the control subjects, our results showed several areas involved in auditory or visually presented language exclusively. In addition, several areas were activated by both tasks. Whereas it is obvious that primary auditory areas will be different with these tasks, the baseline conditions were designed to minimize these activations of the primary auditory and visual cortex. The differential activation patterns during the two language tasks represent basic differences in the processing of language inputs from these different modalities. Data showed that auditory-presented language activates BA 45, whereas the visually presented language task activates both BA 45 and 44. As suggested by the literature on this subject (29–32), a possible function for the Broca area BA 44 is to perform articulatory recoding of written language, essentially a graphene-to-phoneme conversion. A possible function for the Broca area BA 45 is to cope with the demands of either working memory processing or syntactic operations; a common requirement across modalities (3). Both middle and superior left temporal gyri were activated in the VLTask, whereas for the ALTask, bilateral activation of both MTG and STG was observed. This could imply that the inferior part of the STG is used for processing only the auditory phonetic components of language, and because written language also involves phonology, the STG also is activated in the VLTask but only in the dominant STG (2,33–38). The superior temporal gyrus appears to be directly concerned with auditory language input. Converging data from a variety of sources suggest that unimodal auditory systems of the superior temporal gyrus decode the complex acoustic features found in speech, presumably activating neural representations of auditory speech at more-abstract levels (1,39,40). When the higher-order auditory language regions are eliminated from the laterality calculations, the lateralization provided by the auditory task increases, but it is still lower than that found with the visual language task.
The current findings are consistent with the speculation that aspects of IFG, particularly pars triangularis (BA 45), are involved in modality-independent linguistic processing of the auditory and visual sentences, in for example, syntax or working-memory roles. In a recent examination of syntactic/semantic dissociations in sentence processing Dapretto and Bookheimer (3) presented pairs of spoken sentences, and listeners judged whether the two tokens had the same or a different meaning. To isolate syntactic regions, some pairs varied with respect either to form (e.g., active/passive voice) or word order. For the semantic manipulation, tokens differed with respect to lexical–semantic features (e.g., words with same or different meanings in the sentence pairs). Within IFG, although BA 45 (pars triangularis) showed similar activation across conditions, unique foci in BA 44 were obtained in the syntactic condition along with unique foci in BA 47 for the semantic condition. The authors concluded that parts of pars opercularis (BA 44) are implicated when syntactic processing is taxed. These results clearly implicate a unique response in BA44 to the voice and order (syntactic) manipulations. However, it should be noted that because syntactic analysis is required for all sentences, irrespective of condition, the consistent activation of BA 45 (but not BA 44) might be taken to imply, as suggested by the current findings, that aspects of BA 45 serve a functional role in sentence parsing. Recently Caplan et al. in two experiments (11,41) observed maximal response in pars triangularis (BA 45) to manipulations of syntactic complexity. The current results, also implicating a more anterior locus for modality-independent activation, lend support to the argument that BA 45 contains neural systems critical for sentence analysis. Clearly, further studies are required to characterize more precisely the functional differences between these regions in sentence processing and to distinguish syntactic operations from working memory.
The left fusiform gyrus (FuG), which is often referred to as a component of the basal temporal language area (42), was activated in 60% of the subjects, but primarily for the VLTask. This modality dependence is consistent with earlier data implicating the region in visual word-recognition studies (42) and likely reflects this modality-specific processing. Nobre et al. (43) has shown that the anterior FuG is involved in word/nonword tasks and semantic priming effects. Posterior right FuG activation also was activated in our study, probably because of visual shape recognition and letter-string processing (44). We also obtained weak activation in this region in the auditory language task, possibly related to visualization of the auditory input, a phenomenon that has been observed previously in PET studies (9,45).
The patient results are notable in several respects. First, we observed good concordance between fMRI language activation and both Wada language lateralization and intrahemispheric language localization as determined by electrocortical stimulation. This indicates that fMRI activation can provide a useful adjunct to established methods of language mapping in epilepsy patients with dominant hemisphere foci. However, there were some disparities between the techniques that require further discussion, and indicate the need for further refinement of the technique before fMRI can be used more confidently for language mapping in surgical planning.
Second, our results indicated more regions that correlated with the language tasks in the patients as compared with normal controls. This was true for visual language, auditory language, and the modality-independent areas. This has important implications for the understanding of language function in epilepsy patients, and may provide insight into the substrate of language recovery after surgery in the dominant hemisphere. All of the patients had intractable temporal lobe epilepsy, but the group was quite diverse in terms of the exact location of the lesions and the onset of the first epileptic event. The impact of these factors on the patterns of activation for language comprehension requires a much larger sample size and is the subject of ongoing investigation at our institution. Nonetheless, the right-hemisphere shift seems a general characteristic of this group.
One of the important clinical applications of fMRI is the ability to provide noninvasive language lateralization and localization information (46). For the 10 patients that underwent the Wada test, our results showed good concordance with the fMRI in eight of 10 cases. The discordant patients (16 and 17) showed bilateral activation (or weak left lateralization depending on the regions used to calculate laterality) during the fMRI language tasks, but were classified as “left hemisphere dominant” by Wada testing. Although cortical stimulation in these patients identified language areas in the left hemisphere, this does not necessarily preclude the presence of language in the right hemisphere, because the right hemisphere was not stimulated. This apparently discordant finding highlights several important questions with respect to use of fMRI language localization in surgical planning. First, the fMRI and Wada test differ in a fundamental way. Functional MRI is an imaging approach that elucidates the cortex that is differentially active during, and presumably involved in cognitive processing of, the language tasks, but that may not necessarily be critical for language production. In contrast, the Wada test involves pharmacologic deactivation of the language cortex, providing a reversible simulation of the effects of surgery and allowing identification of areas critical for language by producing aphasia during the period of deactivation. The Wada test also typically provides information on the laterality of mnemonic processing that to date has not been easily obtained with fMRI.
The information provided by functional MRI is distinct from the Wada test and cortical stimulation in several respects. The Wada test, although the current gold standard because of its specificity, lacks spatial/anatomic resolution and does not provide the localizing information within a hemisphere that potentially can be obtained using fMRI. In contrast, fMRI has the ability to provide a more detailed intrahemispheric map of language localization with greater functional and anatomic specificity (47). Furthermore, fMRI has the potential to provide information from a single source that is now available only through multiple sources (i.e., Wada testing, cortical mapping). Functional MRI is also noninvasive, and can be more easily repeated than the Wada test and cortical mapping studies with little or no risk to the patient. Thus, its continued development is desirable, and it remains to be seen whether fMRI will serve as an adjunct test procedure to the Wada and cortical stimulation or if, over time, it may partially replace these invasive methods.
The fMRI concordance with intrahemispheric language mapping via cortical stimulation was good, providing both positive (i.e., impairment with stimulation in areas activated for fMRI) and negative (i.e., absence of stimulation-induced impairment in fMRI “silent” areas) concordance in all patients that underwent mapping. Although there are many technical issues to consider when comparing fMRI and cortical stimulation (48), we observed a strong correlation between data from the intra–extraoperative stimulation mapping studies and the functional MRI maps (15 of 15). As expected, cortical stimulation indicated only the strictly sufficient basic language areas for language production (e.g., Broca's area), whereas fMRI highlighted many more areas (71) than did cortical stimulation (15). fMRI showed additional activation in regions such as the SMA, ITG, and deep sulcal regions not only related to productive language areas, but also indicative or more complex language-comprehension processes. The primary limitation of cortical stimulation is lack of brain coverage, which is responsible for the large differences in the number of activation sites observed. In addition, high-intensity cortical stimulation can induce an efficient cortical inhibition or excitation at depth or laterally from the stimulation site, leading to localization problems. Conversely, an insufficient cortical bipolar stimulation intensity (<10 mA–50 Hz) may not provide sufficient stimulation to affect some functional areas. Our experience indicates that fMRI is far more descriptive (in the sense that many more regions of the brain can be explored) and that its sensitivity is higher than that of cortical stimulation. But cortical stimulation is more specific in terms of defining which areas are critical. Functional MRI highlights all regions involved in a specific task, but it does not distinguish between areas involved and areas that are critical to the task. Cortical stimulation conversely (10,42,49) is highly specific for critical productive language areas.
There were several interesting findings in the comparison between the control subjects and patients in this study. First, the strength of the activation was lower in patients. This may be due to drug effects on cerebral activation or blood-flow changes, or possibly due to the effects of the disease itself. Furthermore, patients demonstrated significantly lower lateralization scores compared with normal controls. Total laterality scores between right-handed patients and right-handed controls showed a significant statistical difference (t = 3, p < 0.02). Both the VLTask and the ALTask showed bilateral activation of the STG, MTG, ITG, and IFG in the patients. These differences may provide evidence for area recruitment in epilepsy, but a much larger sample is needed to investigate this issue in detail. These results agree with those of Springer et al. (4), who found higher variability in language dominance for epilepsy patients compared with control subjects when using a simple single-word semantic decision task. Studies of developmental dyslexia have shown reduced laterality in reading disabled children and adults (50,51). An increased role for the nondominant hemisphere in language tasks may also reflect to some degree a compensatory response to impoverished LH language function (52). Behavioral analysis on these patients (Table 5) illustrates they were impaired on various aspects of language tasks. It is possible that the impaired performance on neuropsychological measures of language in these patients also is somehow related to the strength of activation. Correlation of level of impairment with strength of activation will require a much larger sample of patients, as well as broader sampling of language function to address this question adequately.
Table 5. Neuropsychological data
|Patient number||Hand side||Wada test||Boston Naming Test||IQ scores|
|Raw score||z score|| || |
Thulborn et al. (53) recently showed that, in two adult patients recovering from stroke-induced aphasia, fMRI demonstrated modification of the language-activation pattern, at 6 months, from a left to a homologous right hemispheric pattern. An earlier study examining the effects of early unilateral lesions on language laterality found that early left lesions were associated with increased participation of the right hemisphere (54). Comparing left lesion and right lesion subjects, they found no differences in full-scale and verbal IQs and no correlation between either FSIQ or VIQ with rCBF or distributivity measures for listening to sentences (data shown in Table 5). Similarly we also found no relationship between these gross receptive language scores and lateralization, although this is a topic we are currently investigating in more detail.
All of these studies suggest that there may be recruitment of contralateral homologous areas, and hence the development of a contralateral homologous language pattern. In our patients, the etiology of bilateral language is unclear. Patients could have developed either a spontaneous initial bilateral representation of language during childhood or had secondary language bilateralization after becoming left speech dominant with normal language organization during childhood. If patients tend to have more “bilateral” activation as the norm, it will be necessary to determine the extent to which areas in or near activated areas produce impairments if resected. Although one is not likely to resect an area that both the Wada and fMRI studies suggest is involved in language, what is the meaning of resecting and activated area in the “nondominant” hemisphere? Again, further studies are needed to investigate this issue.