New (fMRI) Versus Old (Direct Cortical Stimulation) Technology: Which Is Prime Time for Language Mapping?

Commentary

Language Functional MRI and Direct Cortical Stimulation in Epilepsy Preoperative Planning.

Austermuehle A, Cocjin J, Reynolds R, Agrawal S, Sepeta L, Gaillard WD, Zaghloul KA, Inati S, Theodore WH. Ann Neurol 2017;81:526–537

OBJECTIVE: Presurgical language assessment can help minimize damage to eloquent cortex during resective epilepsy surgery. Two methods for presurgical language mapping are functional MRI (fMRI) and direct cortical stimulation (DCS) of implanted subdural electrodes. We compared fMRI results to DCS to help optimize noninvasive language localization and assess its validity. METHODS: We studied 19 patients referred for presurgical evaluation of drug-resistant epilepsy. Patients completed four language tasks during preoperative fMRI. After subdural electrode implantation, we used DCS to localize language areas. For each stimulation site, we determined whether language positive electrode pairs intersected with significant fMRI activity clusters for language tasks. RESULTS: Sensitivity and specificity depended on electrode region of interest radii and statistical thresholding. For patients with at least one language positive stimulation site, an auditory description decision task provided the best trade-off between sensitivity and specificity. For patients with no language positive stimulation sites, fMRI was a dependable method of excluding eloquent language processing. INTERPRETATION: Language fMRI is an effective tool for determining language lateralization before electrode implantation and is especially useful for excluding unexpected critical language areas. It can help guide subdural electrode implantation and narrow the search for eloquent cortical areas by DCS.

Mapping human brain function began with direct cortical stimulation (DCS) of the brain. In their classic descriptions of functional localization of the cerebral cortex in the 1950s, Wilder Penfield and Herbert Jasper (1) used DCS to map the cerebral cortex of humans, describing many functional areas, including the sensory and motor representations of rolandic cortex, as well as speech and language functional areas. Today, DCS remains the most reliable test for mapping cortical function, with advantages of a very long history of prior experience with the technique. However, DCS has disadvantages, necessitating placement of subdural grid, strip, or depth electrodes, with associated perioperative morbidity and mortality. Another major limitation of DCS is limited sampling of cortical structures. Even the most extensive grid, strip, and depth electrode arrays cover only a fraction of the cortex.

Given the limitations of DCS, the article by Austermuehle et al. covers an important topic in functional mapping, exploring the correlation of DCS with functional MRI (fMRI) for mapping language cortex. Mapping language function with fMRI is noninvasive and samples the entire brain, overcoming the major shortcomings of DCS. For more straightforward functions, such as motor mapping, motor paradigms to obtain fMRI maps are relatively straightforward. However, the optimal tests for mapping language function in fMRI are more complex, reflecting the intrinsic complexity of language function in the brain overall. Therefore, fMRI testing paradigms are less standardized and remain an important topic of study.

Functional MRI paradigms to evaluate language function must account for the complexity of speech and language comprehension. The initial perception of speech engages the bilateral auditory cortex, as well as processing of spectral and temporal speech information, which occur before higher level perception of speech as language. Conceptually, perceived speech eventually activates brain regions that demonstrate long-term representations of language function allowing interpretation of information. The regions of the brain involved in the interpretation of speech and language vary depending on the presentation of information. Therefore, different language paradigms for fMRI protocols show activation in different brain regions. The paradigm used in this study used the typical language fMRI design with language and control conditions, with the language condition theoretically activating language-related regions, and the control condition representing activation of “background” regions. Image processing involves comparison of the language (task) and control conditions, resulting a comparative map of activation of the language condition relative to the control condition.

In a previous study, Binder et al. (2) outlined protocols for mapping speech comprehension areas with fMRI. Language conditions in all protocols involved stimulus of speech. Task state and control conditions varied with each protocol, with variations of passive or active conditions during the task state, as well as variable tasks and conditions during the control conditions. These variations resulted in five different categories of paradigms. Using the concepts of the five different fMRI language protocols, they showed that different fMRI protocols for mapping speech comprehension systems differ dramatically in pattern, extent, and lateralization of regions of activation.

Austermuehle et al. used four different fMRI language protocols: 1) the auditory description decision task (ADDT), 2) auditory categories (AUDCAT), 3) listening, and 4) reading. Given the previously documented variability of regions of activation with different fMRI task paradigms, the authors predictably found broad, differing regions of activation with different protocols. However, to consolidate results, they used standardized objective image processing techniques, including cluster analysis with statistical thresholding (which they defined as P1, P2, and P3), and combining protocols (in different combinations) to include only positive clusters in the combined protocols.

For DCS, the investigators used standard mapping stimulation parameters, and included the following tasks 1) continuous counting or alphabet recitation, 2) visual naming, and 3) reading. Electrode stimulation that reproducibly disrupted either language production or comprehension were defined as language positive. The language positive electrodes were converted to ROIs of 4-, 6-, 8-, and 10-mm radii for comparison with fMRI studies.

Using parameters of 10-mm radius and P2 threshold, the ADDT showed 80.6% sensitivity, 72.7% specificity, 73.4% accuracy. The ADDT showed the best score for the diagnostic odds ratio (DOR, calculated as TP*TN/(FP*FN)) at 11.1. Results from combination masks showed high specificity, with values ranging from 70.6% to 99.1%. The authors concluded that ADDT offers the best balance between sensitivity and specificity. fMRI language maps offer noninvasive methods to assess critical language-related cortical regions. This is especially important in planning intracranial electrode placement in the dominant hemisphere for patients undergoing intracranial EEG for localization of the epileptogenic focus. This study shows fMRI language results can be used to help place intracranial electrodes by excluding unexpected critical language areas before surgery. However, fMRI results do not show high enough specificity to warrant independent localization of language regions without confirmation with DCS.

fMRI brain mapping techniques continue to evolve. There are new techniques for localizing language function, including resting state fMRI. As opposed to task-based functional MRI wherein the subject performs a task which is then compared to a control condition, resting state fMRI evaluates low-frequency fluctuations in the blood oxygen level dependent (BOLD) signal while the subject is at rest. Using this technique, multiple resting state networks have been identified, including the somatosensory, language, and visual networks. Initial results in oncology patients showed this as a potentially useful functional mapping technique (3).

In a recent practice guideline summary for use of fMRI in the presurgical evaluation of patients with epilepsy, Szaflarski et al. (4) concluded that fMRI may be considered for predicting postsurgical language outcomes after anterior temporal lobe resection for treatment of TLE. While this guideline showed that fMRI is useful in evaluating memory and language function, especially as compared to use of the intracarotid amobarbital procedure, the systematic review of fMRI use was hampered by nonstandardized methods for fMRI testing. The issue of nonstandardization of fMRI language mapping paradigms among different studies, as discussed by Austermuehle et al., also limits comparison of studies correlating findings with DCS and fMRI. Given the nonstandardization of fMRI techniques, each institution using fMRI should carefully evaluate its fMRI paradigms and imaging processing procedures to ensure quality studies. Accurate interpretation of fMRI language maps depends heavily on understanding of the functional paradigms and MRI techniques used to derive them.

So, since the 1950s until now, DCS remains the gold standard of language mapping in subjects undergoing evaluation for epilepsy surgery. fMRI remains a useful adjunctive tool for mapping language cortex, especially in planning intracranial EEG electrode placement in patients with suspected dominant hemispheric epileptogenic regions.