Direct electrical stimulation as an input gate into brain functional networks: principles, advantages and limitations
- First Online:
- Cite this article as:
- Mandonnet, E., Winkler, P.A. & Duffau, H. Acta Neurochir (2010) 152: 185. doi:10.1007/s00701-009-0469-0
- 446 Views
While the fundamental and clinical contribution of direct electrical stimulation (DES) of the brain is now well acknowledged, its advantages and limitations have not been re-evaluated for a long time.
Here, we critically review exactly what DES can tell us about cerebral function.
First, we show that DES is highly sensitive for detecting the cortical and axonal eloquent structures. Moreover, DES also provides a unique opportunity to study brain connectivity, since each area responsive to stimulation is in fact an input gate into a large-scale network rather than an isolated discrete functional site. DES, however, also has a limitation: its specificity is suboptimal. Indeed, DES may lead to interpretations that a structure is crucial because of the induction of a transient functional response when stimulated, whereas (1) this effect is caused by the backward spreading of the electro-stimulation along the network to an essential area and/or (2) the stimulated region can be functionally compensated owing to long-term brain plasticity mechanisms.
In brief, although DES is still the gold standard for brain mapping, its combination with new methods such as perioperative neurofunctional imaging and biomathematical modeling is now mandatory, in order to clearly differentiate those networks that are actually indispensable to function from those that can be compensated.
KeywordsBrain mappingConnectivityDirect electrical stimulationNeural networksPlasticity
Understanding the neural and physiological foundations of brain function is one of the most important challenges in the neurosciences. Advances in neurofunctional imaging have made possible better study of the organization of the eloquent structures within the central nervous system. However, the reliability of fMRI is low: reproducibility across multiple fMRI sessions for motor and language testing in healthy volunteers is much less than 100% [33, 69], and in patients with cerebral lesions, cortical activations on fMRI are imperfectly correlated with cortical sites evidenced by stimulations . Furthermore, neurofunctional imaging is not able to differentiate those areas essential for function from those that can be compensated while activated during a task. For instance, it was demonstrated that glioma within the supplementary motor area can be surgically removed without causing any permanent deficit due to the recruitment of the contralateral homologous region, even if this supplementary motor area was detected on preoperative functional MRI in patients performing hand movements . Finally, neurofunctional imaging can map only the grey matter but not the white matter. Despite the development of diffusion tensor imaging (DTI) for tracking axonal pathways, this technique must still be validated, since the results may depend on the biomathematical model used to analyze the MRI data . Furthermore, DTI provides anatomical but not functional information about the white matter tracts.
As a consequence, invasive electrophysiological investigations are currently considered the “gold standard” for brain mapping. First, it is possible to perform extraoperative mapping using a subdural grid. Although this method was extensively advocated in epilepsy surgery, because it also allows detection of the seizure foci, only the cortex can be mapped: it provides no information about the axonal connectivity. This is why in the past decade, ever more authors have advocated to use direct electrical stimulation (DES) intraoperatively, especially in neuro-oncology, since brain glioma invades both cortical and axonal structures [24, 45, 61, 74]. Indeed, DES gives accurate and reliable data on the distribution not only of the cortical eloquent areas, but also of the functional white matter bundles and deep grey nuclei [6, 15, 25, 39]. Undoubtedly, DES has improved the surgical results [18, 71, 76], as regards the maximization of tumor removal as well as the preservation of the patient’s quality of life [19, 72]. In addition, DES has also contributed to a better understanding of the functional organization of the brain .
However, advantages and limitations of this tool have not been reconsidered for a long time, probably because the fundamental and clinical contribution of DES is now well acknowledged. However, such reconsideration is called for in light of the new insights into brain functional organization recently provided by methodological and conceptual progress in the neurosciences. Here, our purpose is to critically review exactly what DES can tell us about cerebral function, that is, the distribution of eloquent regions in the brain, their connectivity and their ability to reorganize—i.e., so-called “brain plasticity.”
An extensive review of the theoretical and experimental studies about the physiological processes underlying cortical electrical stimulation would be beyond the scope of this paper, and we refer the reader to previous works [36, 67, 68]. However, a brief summary of its main principles will be helpful to the understanding of the advantages and limits of this method.
Principles of stimulation: neural membrane dynamics
The membrane potential (MP) of the neuron at rest varies between –60 mV and –100 mV. The principle of electrical stimulation is to generate membrane excitability via an initial phase of passive modification of the local MP at the level of the cathode (i.e., the negative electrode). Here the inner side of the membrane becomes progressively less negative than the outer side. Inversely, at the anode, the MP is further increased to render the target tissue less excitable: the membrane is said to be hyperpolarized. The intensity of this phenomenon depends on the parameters of the stimulations and of the characteristics of the membrane . The outer membrane can be more easily stimulated at the level of the axons rather than that of the cell body [57, 58], and among axons, the depolarization is greater for myelinated ones with a large diameter . If the MP reaches the liminar depolarization threshold, a second phase occurs that begins with the opening of voltage-dependent ionic canals, which allow entry of Na+ ions and which therefore invert the MP between +20 mV and +30 mV. A secondary output of K+ ions, associated with an inhibition of the entering flux of Na+ ions, brings the MP back to its resting state. Once generated, this rapid sequence of MP fluctuation—the action potential—is still the same, whatever the stimulation parameters are (law of “all or nothing”).
Following an action potential, the neural membrane is refractory to all stimuli for 0.6 ms to 2 ms. Thus, myelinated axons produce a single response for each stimulation delivered up to frequencies around 100 Hz . The refractory status precedes a second phase of transitory hyperexcitability, during which the tissue could be stimulated by less intense currents than during the initial stimulus, but with an increased risk of seizure.
Finally, when the neural structures are kept in a state of infra-liminar depolarization, the threshold required to generate the impulse increases: this phenomenon is known as “accommodation.” Accommodation occurs if the MP changes quite progressively, which may be observed when sinusoidal impulses are used. This is the reason why rectangular impulses are recommended for stimulations.
Stimulation parameters: experimental and theoretical approach
The first requirement for stimulation is that it must be entirely safe for the cerebral parenchyma. However, it may induce a lesion by accumulating negative charge at the level of the cathode or by producing metal ions at the level of the anode . The use of biphasic impulses eliminates these risks, since the second stimulus phase reverses the effects of the first. Tissue damage can also be caused, on the one hand, by excessive heat, produced specifically by hydrolysis, which could cause vacuolization and chromatolysis, and on the other hand, by “leakage” of the intracellular current, which goes from the anode to the cathode through the cytoplasm, posing a risk of lesions on the mitochondria and the endoplasmic reticulum, or even by alteration of the homeostasis when the neurons are activated in a manner that is too repetitive and synchronous . These risks are linked to the density of the charge, while the relatively brief and intermittent periods utilized for human stimulation testing do not appear to cause structural damage at the light microscopic level at charge densities that exceed the threshold for damage established in animal studies with more continuous, chronic stimulation schedules .
The second requirement for this stimulation is that it must induce a reproducible response when applied to neural structures. The relationship between stimulation parameters and tissue response can be summed up by the intensity/duration curve, where the current intensity needed to produce a response is plotted as a function of the duration of the impulse . The rheobasis is thus defined as the minimal intensity required to generate an action potential at elevated values of impulse duration. Chronaxie is the impulse duration required to induce a response when the stimulus intensity is twice the rheobasis. The chronaxie corresponds to the point on the intensity/duration curve where the energy allowing an action potential to be generated is minimal for a charge twice the minimum charge: this is the optimal stimulation point (best benefit/risk ratio) .
The chronaxie point depends on the membrane characteristics. Indeed, chronaxie can be significantly modified by the degree of cerebral maturation, notably by myelinization: chronaxie of non-myelinated nerve fibers is considerably longer (0.4– 3.5 ms) than that of the myelinated axons (0.05–0.4 ms) . The size of the fiber also seems to be a factor, in that the axons of greater diameter are more readily excited .
Apart from membrane properties, the chronaxie is triggered by the activating function, which is the second spatial derivative of the voltage seen by the membrane [80, 82]. The activating function results from the complex interplay between stimulation parameters (intensity), electrode design (mono/bipolar, size and geometry, orientation with respect to the fibers) and tissue characteristics (impedance).
Regarding impedances, the few studies on this topic  have produced resistance values of 250 Ohms for gray matter, 500 Ohms for white matter and 65 Ohms for cerebrospinal fluid . Moreover, impedances can be modified by the state of the patients (awake or under general anesthesia). Finally, any pathological process, whether lesional (tumor) or non-lesional (epilepsy, post-ictal status), can interfere directly with the tissue’s excitability .
Although two electrodes are still required to produce a current, the stimulations are considered to be “monopolar” if only one of the electrodes is “active” (in general, the cathode), i.e., localized in relation to the target tissue, while the reference electrode (in general, the anode) is located at a distance . Whereas the current density is distributed in a relatively uniform manner around the electrode, each tissue located in the current’s pathway can nevertheless be stimulated, especially if its depolarization threshold is less than the target threshold. To reduce this risk of a false positive, it is preferable to use a bipolar stimulation, i.e., where the cathode and the anode are both “active,” or in other words, both are located at the level of the target tissue . Only structures located between the two electrodes are stimulated, and there is less risk of diffusion and therefore a greater precision . It is worth noting that the current density is highest between the two electrode contacts, an essential fact to understand especially for the use of strip and grid electrodes.
Finally, the size and geometry of the electrodes tips strongly influence the effective stimulus volume . As a consequence, different levels of the map could be observed when using a mapping function with a different electrode design (e.g., microstimulator versus Ojeman bipolar).
Stimulation parameters: practical intraoperative approach
In practice, bipolar electrode tips that are spaced 5 mm apart and deliver a biphasic current (pulse frequency 60 Hz, single-pulse duration 1 ms and intensity from 6 to 18 mA under general anesthesia or from 2 to 6 mA under local anesthesia) are applied to the cortex . DES allows the mapping of motor function (by inducing involuntary motor response if the stimulation is applied at the level of a primary motor site, even under general anesthesia), somatosensory function (by eliciting dysesthesia described by the patient himself intraoperatively) and also cognitive functions such as language (spontaneous speech, picture naming, comprehension, reading, writing, bilingualism, etc.), calculation , memory  or even visuo-spatial processing [20, 81]. In these cases, patients are awake, and transient disturbances are generated by applying DES at the level of a functional “epicenter” . Functional disturbances should be analyzed by a dedicated person (nurse, anesthesiologist, neuropsychologist and optimally a speech therapist) in order to accurately interpret the kind of disorders induced by DES, for instance, speech arrest, anarthria, speech apraxia, phonological disturbances, semantic paraphasia, perseveration, anomia, dyscalculia, etc. Thus, DES is able to identify in real-time the cortical sites essential for the function (i.e., they must be preserved), before beginning the resection, in order to both select the best surgical approach and to define the cortical limits of this resection .
In addition, DES can also identify and preserve axonal pathways crucial for sensorimotor, language and other cognitive functions [11, 15–17, 81]. Indeed, it allows the study of the anatomo-functional connectivity by directly and regularly stimulating the white matter tracts throughout the resection and by eliciting a functional response when in contact with the essential eloquent fibers, according to the same principle as that described at the cortical level. Thus, except for motor mapping, it is important that the patient remains awake during the entire removal of the glioma and not only for the initial stage of cortical stimulation. This is why intraoperative stimulations are time-consuming, and the number of tasks during surgery is limited by the progressive tiredness of the patient.
Stimulations may also generate seizures. To avoid this drawback in awake patient, it is recommended to use intra-stimulation electrocorticographic recording, so that the threshold at the origin of “after-discharge” occurrences is never exceeded (for a review, see )—except in children due to non-myelinization of the fibers . It is worth noting that there is no need to perform electrocorticographic monitoring to detect after-discharge potentials for motor mapping . Moreover, in cases of stimulation-induced seizures, the use of cold Ringer’s lactate is recommended to abrogate the seizure activity .
Advantages of DES: An input gate into the funtional networks
One of the major advantages of DES for brain mapping in adult patients is that they intrinsically do not cause any false negatives—if nevertheless the methodology is rigorously applied (see below). Indeed, each eloquent structure, whatever its actual role in brain function, will be in essence electrically disturbed by DES, which necessarily will induce a functional consequence. This optimal sensitivity explains why DES is currently considered the gold standard in brain mapping. Indeed, this is the most important point for neurosurgeons, since intraoperative DES enables them to tailor the resection according to functional boundaries—thus minimizing the risk of permanent deficit . Moreover, this advantage also explains why DES is used to validate the non-invasive method of neurofunctional imaging (fMRI, PETscan and MEG) [41, 43, 70] as well as the recent technique of DTI .
Regarding the spatial resolution, while it has been argued that the currents delivered by the tips of the bipolar electrodes can diffuse over the cortex, thus limiting the spatial accuracy of the tested areas, it is necessary to distinguish two types of "diffusion." The first one involves the local spread of the depolarizing electric field over the cortical surface (either by passive diffusion or by local synapses). Using optical imaging, it was shown more than 10 years ago that the spatial area affected by this spreading is about 5 mm in diameter . The diffusion has not been investigated experimentally for axonal stimulations, but theoretically, it should be of the same order, or even smaller when the bipolar electrode is parallel to the direction of the fasciculus. Thus, the spatial resolution of DES is about 5 mm, namely less than the 1 cm spacing of the subdural strip and grid electrodes used for recording epileptiform activity and for cortical mapping before epilepsy surgery. In a large series of patients who underwent invasive investigations, we compared the extraoperative cortex stimulation with the intraoperative cortex stimulation. In many patients, we observed a marked discrepancy in stimulation responses, even when exactly the same stimulation parameters were used. For example, we registered a “displacement” of the central sulcus in a patient with perirolandic epilepsy. In other patients, we found different results by stimulating the speech area. In some of our cases, the additional intraoperative DES, used in a systematic way in our epilepsy surgery program, allowed a more aggressive approach and the most radical excision of the epileptogenic lesion, whereas in other cases, the additional intraoperative DES minimized the risk of permanent postoperative aphasia. In a third, smaller number of patients, we found slight differences between the two methods. These preliminary results indicate that the disadvantage of the width distance between the stimulation points of the grids (regularly 1 cm) should be neutralized thanks to the spatial resolution of intraoperative DES.
The second type of "diffusion" refers to the biological and physiological propagation of the stimulation. Indeed, when stimulating the neuron (whatever the stimulated part), the action potential propragates by physiological conduction along the axon, and consequently, neurons at the end of the axons can be stimulated by these presynaptic currents. Similarly, when stimulating axonally, the action potential propagates both forward (in a physiological way) and backward (in a non-physiological way). So, in essence, DESs are highly non-local: they enter the whole network that sustains a function. The stimulated point (axonal or cortical) is only an input gate into this whole network. In other words, one should forget about localizationist models as well as a connectionist approach and shift toward a non-local theory when analyzing how DES works.
It is worth noting that the perturbation induced by DES in a functional network is small enough that it propagates only in a “sub-circuit,” thus inhibiting solely a specific component of the tested function. This is why for cortical mapping of language function, depending on which “sub-circuit” is disturbed, one may observe phonological errors, speech apraxia, semantic paraphasias, anomia or syntactic mistakes. Of course, when increasing the stimulation intensity, one ultimately generates a speech arrest . Similarly, when stimulating axonally, one generates a dysfunction of a specific network. While in a connectionist point of view this dysfunction is mediated by a mere transient disconnection between two still effective areas (i.e., only the link is disturbed, corresponding to a connectionist point of view), the non-local theory assumes that one or both of the areas linked by the stimulated pathway will be disturbed. Anyway, the resulting effect mimics disconnection syndrome, as observed in lesions of white matter. For instance, DESs induce conduction aphasia during stimulation of the left arcuate fasciculus , semantic disturbances during stimulation of the left inferior occipito-frontal fasciculus  or transcortical motor aphasia during stimulation of the left dominant subcallosal medialis fasciculus, which connects the supplementary motor area and the cingulum with the head of the caudate nucleus . Finally, keeping in mind this unifying concept of network stimulation, one can better understand why conduction aphasia can be observed for axonal  as well as cortical stimulation [64, 65]: wherever the stimulation, the same network is disrupted.
Since DES allows the performance of on-line anatomo-functional correlations at both the cortical and axonal levels, and since each eloquent discrete site identified is only an input gate to a wider functional network, DES is consequently a perfect tool for the study of spatio-temporal brain connectivity. Indeed, it is now well known that to generate a function, it is necessary that several areas work together. Specific temporal dynamics allow synchronization within the distributed networks [3, 5, 50]. DES offers the unique opportunity to directly access these dynamics by associating electrocorticographic recordings to the stimulation. Such methodology has been applied using implanted grids for pre-surgical planning of drug-resistant epileptic patients. It has been shown that the stimulation of a cortical site induced a signal in a distant (but axonally linked) area, i.e., the cortico-cortical evoked potential (CCEP) [34, 48, 49]. Precise analysis of the delay between stimulus and the peak N1 of the CCEP is blurred by the so-called direct negative cortical response . Intraoperative axonal stimulations would be an interesting tool to overcome this limitation, allowing to study both forward and backward axono-cortical-evoked potentials, thus giving insight into the computations performed by the two distal cortical areas on both ends of the pathway.
Finally, this paradigm shift is also ongoing in other fields of functional neurosurgery, like deep brain stimulation or epilepsy. First, it is now widely accepted that stimulation of the subthalamic nucleus also interacts with the whole network of the basal ganglia [51–54] (including the primary motor cortex and its feedback). From a practical point of view, this understanding has suggested new targets in neuromodulation for Parkinson’s disease, such as the primary motor area [8, 60] or even the spinal cord . Similarly, there is now some evidence that some epilepsies are a “network disease” rather than a focal one [4, 62, 77].
Finally, diffusion within a large-scale network may also represent a limitation of DES, as is discussed below.
Limitations of DES
It is important to stress that the slightest technical approximation can result in false negatives. Indeed, we have previously explained in detail that an intensity of stimulation that is too low, lasts too short a time or is performed during a transient post-epileptic refractory phase may lead to an erroneous “negative mapping” . Nevertheless, such failure can be avoided by strictly following the theoretical and practical rules of stimulation—except in children, when fibers are not yet myelinated .
However, it was previously mentioned that stimulation excites the largest size of axons and neurons. In the voluntary firing of the neurons (such as voluntary movements) in case of anterior horn cells in the spinal horn, due to the size principle, not necessarily the largest neurons but rather small neurons start firing. The same situation would occur, namely, external cortical stimulation by an electrical pulse excites the largest neurons, which are not necessarily mainly involved in the voluntary task performances of humans. It may explain some of the differences in the results between DES and the final outcome of resective surgery.
A more subtle limitation would be an inappropriate functional testing. For example, if intraoperative testing is only based on a visual naming task, trouble with finding words can arise after resection of the posterior superior temporal gyrus because of the anatomic dissociation of visual and auditory naming [29–32]. In the same way, it has been recently shown that repetition tasks should be added for lesions involving this same region [64, 65]. Moreover, it has been recently shown that patients who underwent operations for a tumor within the left dominant hemisphere may exhibit a postoperative working memory deficit, despite extensive intraoperative language mapping [9, 79]. This is a result of the unspecific mapping of working memory by adapted tasks during surgery. Thus, the erroneous conclusion is drawn that the tissue was “not functional”—which may have been true for language. It was also recently demonstrated that specific testing of spatial awareness was mandatory during surgery within the right “non-dominant” hemisphere in order to avoid any postoperative hemineglect . Thus, the limited number of feasible peroperative tasks can limit the sensitivity of function detection. This is why it is important to optimize the selection of the intrasurgical tests for each patient on the basis of the preoperative functional assessment with both extensive neurological and neuropsychological examinations, as well as neurofunctional imaging [14, 79].
"Acute" false positives
While the sensitivity of intraoperative electrical mapping is very high, its specificity remains a matter of debate.
First, the patient’s tiredness after approximately 2 h of continuous functional assessment during an awake procedure may cause the accuracy and the rapidity of the answers to decline. Thus, the resection may have to be temporarily interrupted, despite the presence of a speech therapist, who analyzes the response intraoperatively. When, once rested, the patient is able to perform the tasks again correctly, then the surgeon has to perform the electrical mapping again. The goal is to differentiate between the immediate proximity of the eloquent structures (transient deficit during DES, leading to cessation of surgery) and tiredness of the patient (no disturbance during stimulation, thus continuation of the resection).
Second, DES may induce partial seizures that can look like a “positive effect” of the stimulation. For instance, a partial seizure elicited by DES may generate a transient language disorder after stimulation within the dominant hemisphere, leaving the wrong impression that this area is crucial for the function. Nevertheless, a rigorous methodology of stimulation, as detailed above, combined with electrocorticographic recording during the electrical mapping—in order to detect both afterdischarges with the goal to find the optimal threshold of stimulation as well as any seizure during DES—will avoid such false positives. Above all, the reproducibility of the symptoms elicited by DES during the initial cortical mapping should be confirmed by the subcortical mapping. The coherency of the two maps will guarantee that the organization of the network (and not only of isolated epicenters) has been correctly understood [11, 15].
Last but not least, while there is no cortical spreading of DES, there is a propagation of the stimulation along the axon within a network wider than the sole area tested. Although this property allows the study of the connectivity with a very high sensitivity, it can also lead to false positives. Indeed, it has been reported that when electrical mapping was performed using subdural grids for preoperative planning in chronic epilepsy, DES of a cortical site elicited a signal in a remote but axonally linked region . Such mechanisms, especially if the stimulation generates a backward propagation, might contribute to explaining at least a part of the observations that resections of “positive” areas (i.e., with functional disturbances induced when stimulated) did not cause postoperative deficits—e.g., the basal temporal area was removed by some authors without permanent sequelae . Furthermore, it is not known how many synapses belonging to a functional network DES actually diffuse through. This is true for both upward and backward propagation. This non-specificity may be a serious limitation of DES, since it can lead to the premature interruption of the resection because of “false” essential functional boundaries.
“Chronic” false positives
Finally, another limitation of DES is the valid identification of eloquent areas during surgery, which, however, could be functionally compensated. This can be the case with the fusiform gyrus, which some authors decided to resect despite the induction of language disturbances during DES . The transient worsening in several patients supports the real functional role of this structure in these cases, whereas no long-term deficit was noted. The same observation has also been made for the supplementary motor area. Indeed, although its stimulation can induce motor and even language responses, it is nonetheless possible to remove this area, causing a transient “supplementary motor area syndrome” followed by a complete recovery—which is especially due to the recruitment of the contralateral homologue area . Such functional compensation is explained by mechanisms of brain plasticity, which can be at least partly facilitated by the surgical resection itself . In a recent paper, Seeck et al.  presented a patient operated on twice for epilepsy. The second resection, performed 8 months after the first one, involved an area in which stimulation had induced speech arrest during the investigation prior to the first intervention. No postoperative worsening was observed. It is our opinion that long-term plasticity is the major explanation, as we previously described in low-grade glioma . Thus, an “essential” area at the time of surgery is not systematically a “definitive” crucial area. The next step is to try to predict the individual plastic potential in order to plan multiple-stage surgery, guided by repeated mappings and with the goal of extending the quality of resection while preserving the quality of life thanks to a better understanding of the abilities of functional reshaping in each patient .
If its rules of use are rigorously applied, the sensitivity of DES for detection of cortical and axonal eloquent structures is 100%. This is the reason why DES is still considered the gold standard for cerebral mapping. The main disadvantage of this technique is its suboptimal specificity. DES may indeed indicate that a structure is crucial due to the induction of a transient functional response when stimulated, whereas (1) this effect is generated by the spreading of the electro-stimulation, antidromically along the axons to a really essential network and/or (2) the region stimulated can be functionally compensated owing to brain plasticity mechanisms. In other words, at least in a few cases, DES could wrongly lead to a premature interruption of the resection. However, it should be kept in mind that surgical resection performed too close to structures responsive to DES may generate permanent deficits . Since the primary goal of medical doctors, and especially neurosurgeons, is “primum non nocere,” it seems reasonable to interrupt the surgery according to the functional boundaries provided by DES.
Longitudinal studies using non-invasive neurofunctional imaging  will lead to a better understanding of the brain’s abilities to reorganize after surgical resection performed under DES. This could allow in some well-defined situations to resect some areas responsive to DES (as in the case of the supplementary motor area).
DES could be a powerful tool to study brain effective connectivity. Indeed, each area that is responsive to stimulation is in fact an input gate to a large-scale cortico-cortical and cortico-subcortical network rather than an isolated discrete functional site. Recording the distant effects of an axonal stimulation [48, 49] will allow to infer the anatomical substrate of edges in a graph analysis of a network  as well as to refine our knowledge of its temporal dynamics.
Biomathematical modeling will certainly help to answer the question “How does cortical and axonal stimulations work?” Such modeling is expected to give a better description of the neurophysiological mechanisms involved both at the scale of single cells and long-range networks. Such modeling is under way for cases of deep brain stimulation  and cortical stimulation for pain .
Actually, we concur with the conclusion proposed by Agarwal et al. : “We are confident that by continuing to study electrical stimulation modeling and comparing it to real patient deficit data, we will have a better understanding of the speech areas of the human brain,” but we would replace “areas” with “networks.”
The authors thank Dr. Mélanie Pelegrini-Issac and Mrs. Judy Benson for helpful comments on the manuscript and English revision.