Keywords

19.1 Introduction

Neurosurgery aims to maximize the resection of brain tumors while minimizing involvement of eloquent cortex, thus improving survival while preventing lasting neurological deficits. Locating eloquent cortex presents a number of challenges: the unique anatomy of individual brains, tumor-induced distortion, and cortical reorganization render individual preoperative brain mapping essential [1, 2]. Current gold standards for brain mapping are highly invasive and consist of intraoperative electrocortical stimulation (ECS) [3] and the preoperative intracarotid amobarbital Wada test [4]—these techniques are by no means perfect and entail their own set of limitations. ECS, for example, is limited to the site of the craniotomy and to the superficial cortex exposed, as deeper sulci are poorly accessible. Functional magnetic resonance imaging (fMRI) [5, 6], on the other hand, combines whole-brain mapping with the safety of a non-invasive procedure and has been validated against ECS and Wada for motor mapping, language lateralization, and to a lesser extent language localization [7,8,9,10,11,12,13,14,15,16,17,18]. Furthermore, fMRI allows for assessment of operative feasibility, stratification of operative risk, guidance of surgical planning in terms of approach and craniotomy size, and selection of patients for ECS before an invasive procedure has been carried out.

Preoperative fMRI evaluates tumor resectability in terms of lesion-to-activation distance (LAD) for motor and language networks as well as language lateralization. In a surgical study comparing fMRI activation to ECS, resections with an LAD of less than 1 cm resulted in 50% postoperative deficits, dropping to 33% between 1 and 2 cm, with no deficits reported in LADs greater than 2 cm [7]. The LAD based on fMRI, however, cannot be regarded as absolute—depending on statistical parameters (thresholds) defined by the user fMRI activated cortical areas will increase or decrease in size; a further study recommends ECS be performed within an fMRI-based LAD of 10 mm [19]. Language lateralization has implications in patients with tumors and epilepsy, especially for pathologies in the temporal lobe. Furthermore, fMRI should be combined with diffusion tensor imaging (DTI) whenever possible, as the preoperative evaluation of underlying white matter tracts is imperative. Still, fMRI has been shown to alter neurosurgical planning in up to 49% of cases [20] while decreasing craniotomy size in 15% cases. fMRI data can also be implemented into neuronavigation systems [21,22,23], but a brain shift of up to 20 mm can occur on craniotomy [24] and prediction of brain shift is impossible [25].

Presurgical motor fMRI is recommended in patients where tumor invasion of the precentral or postcentral gyrus is suspected or when structural landmarks are no longer distinguishable on anatomical MRI sequences [26]. For language fMRI, indications include patients presenting with tumor-induced language deficits. Keeping in mind the reported left hemispheric dominance in language processing (95% of right-handed individuals and 70% of left-handed individuals) [27], patients with left hemispheric tumors suspected of being in close proximity to essential language areas should be considered for presurgical fMRI. Furthermore, left-handed patients, especially those with right-sided brain tumors, should be considered to assess for atypical language dominance. This chapter focusses on task-based functional MRI only, in the preoperative setting in patients with brain tumors or epilepsy, accounting for the best established and validated clinical application of fMRI. The application of resting-state fMRI as a clinical diagnostic tool is not validated yet and therefore not addressed in this chapter.

19.2 From Brain Physiology to BOLD fMRI Signal

Functional magnetic resonance is a non-invasive imaging tool making use of magnetic changes induced by hemodynamic changes associated with brain activity. Activated neurons require a higher influx of metabolic substrates to meet increased energy and oxygen demands. The brain’s response is to dilate local supplying arteries and capillaries for a greater blood inflow. For reasons not fully understood, the brain’s response is over-proportionately strong [28], resulting in a washout of the venous deoxyhemoglobin and a shift in the local (venous) ratio of blood in favor of oxyhemoglobin. This venous-based effect is the basis for the blood oxygen level-dependent (BOLD) fMRI signal. Paramagnetic deoxyhemoglobin results in magnetic field distortions and lower signal on MRI; its washout with diamagnetic oxyhemoglobin results in a signal increase [28]. The change in magnetism due to altered blood flow can be detected on MRI using pulse sequences designed to highlight these magnetic differences—generally, T2∗—echo-planar imaging (EPI) [29, 30] is used. Brain function on fMRI is therefore mapped using regional blood flow changes as surrogates of neuronal activity through a process known as neurovascular coupling [31] thought to be realized by astrocytes [32, 33]. Pathologies such as tumors and arteriovenous malformations (AVMs) can disrupt this process resulting in neurovascular uncoupling.

Vasodilatation and vasoconstriction are lethargic processes compared to the rapid response of neurons. This vascular response is known as the hemodynamic response function (HRF), showing an initial dip before increasing to a higher value and returning to baseline after 16–20 seconds [34, 35]. The signal from an fMRI experiment is, therefore, one of relatively low frequency correlating to these sluggish vascular changes.

For task-based activation studies, two major experimental designs exist to map brain function with fMRI; they are known as block designs and event-related designs. The block design basically consists of two alternating continuous conditions (or epochs) of predefined duration, one of which represents the experimental task (such as repetitive finger tapping) with the other representing control (e.g. rest). This creates differing brain activation states which are then statistically compared to each other. Block designs produce the larger relative BOLD signal change on fMRI [36], have increased statistical power [37], and are traditionally used in the preoperative setting. The event-related (ER) design, on the other hand, involves series of single stimuli being presented at random, reducing adaptation effects by the subject [38], and is preferential when higher temporal resolutions are required [39]. Due to the higher demands and difficult applicability in the clinical setting, ER-fMRI has its domain more in neuroscientific functional imaging research.

A high signal to noise ratio (SNR) is essential to any imaging study, especially when it comes to diagnostic use in individual patients. However, on fMRI, noise can be even greater than the signal acquired with magnitudes of up to 2–4% [40] and represents a major challenge to a successful fMRI study. Major physical sources of noise include thermal noise from the subject as well as cardiac and respiratory noise [41] as well as variations in baseline neural metabolism. On fMRI, noise is increasingly pronounced at low frequencies [42] on the order of those alluded to in the HRF. Fortunately, fMRI signal is highest when low frequency changes occur [43, 44]. Power spectrum analysis comparing signal (from the HRF) to the noise reveals ideal epoch lengths of 15.5 s [45]. The block design fMRI experiment is therefore delicately planned into paradigms using epochs between 15 to 30 s in duration [46] traditionally run for around 4 min [41]; in our experience, the duration of a complete clinical fMRI examination consisting of multiple paradigms ranges between 10 and 20 min [47].

19.2.1 Preprocessing the Acquired 4D Data

Various problems arise when scanning a spherical, moving object with intrinsic inhomogeneities (such as a human’s head) inside a magnetic tube. Here, proper preprocessing including artifact corrections of the acquired data is of utmost importance.

The combination of EPI sequences with magnetic field inhomogeneities caused by air-tissue interfaces results in geometric distortion. Image distortion correction is therefore required as images would otherwise suffer from stretching, pulling, and signal dropout. In comparison to anatomical images, displacements of 10 to 20 mm have been described in the functional data set, highlighting the importance of this data processing step [48,49,50].

Second, the 3D volume of the brain is scanned slice by slice at slightly different time points, either from top to bottom, vice versa or in an interleaved fashion. The functional volume, however, is considered to be a single time point, as differences in regional blood flow across two (or more) neighboring slices occur simultaneously. To correct for these time differences, which can be up to several seconds [51], slice timing correction to a fixed time point is used.

Subsequently, correction for patient motion must be performed. Head motion and rotation are possible in all three axes (X, Y, Z); therefore at least a six-parameter motion correction is warranted. Similar to slice acquisition correction, a reference is created for which all other displacements are aligned. Not all effects of patient motion can be undone as the magnetic field is distorted through motion. For this reason, we recommend data sets exceeding 2 mm of head motion to be rejected (rule of thumb: reject more than half the voxel size).

Finally, voxels are spatially smoothed by convolution with a 3D Gaussian kernel, improving signal to noise ratio by creating weighted averages with neighboring voxels. Statistical power is improved and the likelihood of obtaining significant results increases, all at the expense of spatial resolution.

Complex statistical modeling, for example, using the general linear model (GLM) [52,53,54] is implemented to reveal the vital yet minor BOLD signal changes between the two activation states, which can be as low as 1–5% [55]. A discussion on the GLM is beyond the scope of this chapter, the interested reader is referred to the references provided.

Before reading, the functional data set must be overlaid onto anatomical images (e.g. a 3D-T1 weighted data set, 1 mm isotropic, sagittal acquisition) using dedicated software to determine the anatomical correlates of the different functional activations. Next, the statistical threshold used for interpretation of the functional data is set—this step is user-dependent and often of debate. We recommend a dynamic thresholding procedure when performing clinical fMRI studies on individual patients for diagnostic medical use [56,57,58,59,60] to localize the anatomical correlates of each activation precisely. Keep in mind we are locating the “center of gravity” (COG) of the activation clusters; thresholding (step-by-step instructions described later) results in COG growth (when reducing the statistical threshold) or shrinkage of the COG activation (when increasing the statistical threshold) and by no means represents true physiological boundaries of the activated cortex (Fig. 19.1).

Fig. 19.1
figure 1

Changing the statistical threshold changes cluster size, using the motor network as an example in a patient with a left-sided tumor of the pre-and postcentral gyrus. Paradigm: finger tapping of the right hand. The statistical threshold (False Detection Rate FDR) increases from lowest (left) to highest (right)—the two images on the right use the statistically more stringent Bonferroni method. Motor areas: (1) primary motor area M1/S1, (2) premotor cortex (PMC), (3) supplementary motor area (SMA), (4) superior parietal lobule (SPL), (5) co-activation of the primary motor area ipsilateral to movement

Full size image

19.3 Thresholding: How We Do It

19.3.1 Thresholding

  1. 1.

    First, set the false discovery rate (FDR) to 0.05.

  2. 2.

    Increase the threshold so that no functional activation is displayed. Gradually lower the threshold to the first cluster of activation with a predefined volume (our standard is 36 mm2).

  3. 3.

    Checkpoint: Does the cluster we see actually represent an expected hemodynamic response function (HRF)? Signal characteristics for this cluster are analyzed using the signal time course. The correlation coefficient (r) of the signal time course for BOLD signal is compared to the modeled HRF. A minimal correlation (r) of >0.40 is recommended with p < 0.05 (Bonferroni corrected) for cognitive paradigms and r > 0.6 for motor paradigms. Unphysiologically high BOLD signal changes (ΔS%) over 6% at 3 T—or 3% at 1.5 T respectively—are suspicious for a dominant venous origin (Fig. 19.2).

  4. 4.

    Continue lowering the threshold until activations are identified in all diagnostically relevant ROIs.

Fig. 19.2
figure 2

Artifacts: On the left, an example of motion-induced artifacts pronounced directly underneath the skull in a convex shape best appreciated in the coronal plane. In the bottom left corner, the corresponding motion correction parameter graph over time. On the right an example of a pulsating vein, in this case the vein of Galen. While the correlation of the BOLD activation to the modeled hemodynamic response function is high (r = 0.85), the mean signal change of 7.55% is above the acceptable limit—for this reason, this area of BOLD activation must be deemed an artifact

Full size image

19.4 Motor and Sensory fMRI

19.4.1 Introduction

fMRI for localization of brain motor areas has been validated in multiple studies but ultimately does not fully replace the gold standard ECS in tumor surgery. Early studies comparing fMRI to ECS reported accuracies in locating important motor areas of 100% [7, 8]; recent literature, however, reports accuracies at 91% [9] with sensitivities ranging between 85% and 97% [10,11,12]. Mapping of functional motor areas is generally accomplished by asking the patient to perform voluntary muscle movements such as finger tapping, tongue movements or lip pouting, and toe flexion/extension; this way, three separate areas on the primary motor cortex can be located (Fig. 19.3). Supervised patient training before the examination is essential. The following section is a summary and discussion of functional motor areas with practical tips for unusual situations.

Fig. 19.3
figure 3

BOLD activation mapping of the motor network in a patient with a left-sided tumor of the pre- and postcentral gyrus: (1) Primary motor area M1/S1, (2) Premotor cortex (PMC), (3) Supplementary motor area (SMA), (4) Superior parietal lobule (SPL), (5) Co-activation of the primary motor area ipsilateral to movement. Images on right: tongue (left), hand (middle), and foot (right). Signal-time courses of the BOLD activation (red) compared to the modelled hemodynamic response function (HRF, grey) for tongue (left), hand (middle), and foot (right)

Full size image

19.4.2 Cortical Motor Areas

19.4.2.1 Primary Motor Area (M1)

The primary motor cortex is found on the precentral gyrus and is arranged in a distorted, inverted body representation known as a homunculus—feet and toes are located dorsomedially near the central sulcus (paracentral lobule), the M1-hand representation is located further down the hemispheric surface in an anatomically distinct area known as the “hand knob” [61] and the face is represented on the lower hemispheric convexity. Focal defects of M1 result in complete paralysis of the contralateral muscle group with a limited chance of gradual recovery of proximal functions, but lasting deficits in fine motor control [62]. Self-triggered movements of the tongue or lips, hand or fingers, and foot or toes are used to locate M1 on fMRI. Unilateral finger tapping activates the contralateral primary motor cortex as well as the premotor (PMA) and supplementary motor areas (SMA) in both hemispheres. Activation of the corresponding primary somatosensory (S1) somatotopic body representation of the postcentral gyrus is also intrinsically associated with voluntary movements. As a consequence, such fMRI BOLD-responses reflect the so-called primary sensorimotor activation. In cases where only primary sensorimotor cortex fMRI localization is desired, alternating movements of the right and left side of the body will lead to primary sensorimotor activation in both hemispheres without visualization of secondary cortical areas. This is due to the fact that the movements of either side activate the secondary areas (PMA, SMA) of the cortical motor network bilaterally. Such continuous activation leads to insignificant differences on statistical analysis, resulting in non-visualization on fMRI [63]. This is advantageous in patients with greatly distorted anatomy where undesired activation of the secondary motor areas could interfere with the identification of the primary motor area. A word of caution with this technique is warranted as bilateral activation of the primary motor cortex upon unilateral voluntary movements has been described [56]. On the contrary, bilateral motor activation should be avoided if cortical reorganization needs to be addressed in clinical fMRI. Here, strictly unilateral movements are utilized to thoroughly study changes in primary and secondary cortical motor areas. As mentioned earlier, this is not possible when employing bilateral movements for motor cortex activation.

Patients with lesions involving eloquent motor areas are sometimes unable to perform fine motor movements due to paresis. In these cases, finger tapping can be replaced by larger movements such as hand clenching. Generally, movements involving larger muscle groups are at higher risk for inducing motion artifacts, so the patient must be focused on keeping their head still. Options in fully paralytic patients include imagination of movements, showing activation of primary [57] and secondary motor areas [64], and passive motor movements [65] performed with the aid of a second individual (Fig. 19.4). Indirect mapping of the primary motor cortex through localization of the primary somatosensory cortex (S1) is also possible. Various stimuli such as fully automated pneumatically driven tactile stimulation [58, 59, 66] and manual stroking [63, 67] can map somatosensory cortex on the postcentral gyrus, obtaining indirect information on the location of the precentral gyrus [63]. Of note, tactile stimulation also elicits bilateral secondary (SII), integrative sensory areas [68].

Fig. 19.4
figure 4

Active and passive movements for motor mapping: left row: BOLD activations can be seen in a volunteer actively moving their hand (above) and foot (below). Middle: passive movements of hand (above) and foot (below) performed by a second individual in the same volunteer. Right: BOLD activations of passive movement of hand (above) and foot (below) in a highly uncoordinated patient, again performed by a second individual

Full size image

19.4.2.2 Premotor Cortex (PMC)

The PMC is located anterior to the primary motor cortex on the anterior aspect of the precentral gyrus and posterior aspect of the middle frontal gyrus. The PMC appears to house coordinated, complex movements responding to visual cues [69], is activated bilaterally during unilateral muscle movement, and shows higher activations with increasingly complex movements [70]. These observations can be advantageous for locating the premotor cortex using the opposite limb [60] when voluntary movement of the contralateral limb is not possible. The PMC can also be activated during imagined complex movements [70] and while observing actions [71] without subsequent muscle movement. On postoperative resection, transient apraxia has been reported, with lasting deficits being attributed to subcortical lesions [72]. Naming deficits have also been described [73] on ECS which is elaborated in the language section.

19.4.2.3 Supplementary Motor Area (SMA)

The supplementary motor area is located on the medial aspect of the posterior superior frontal gyrus, essentially in continuation with the precentral gyrus located further laterally. The anterior SMA is involved in initiation, planning and temporal sequence selection of voluntary movements [74] and retrieval of motor memory [75], housing memorized programs, and has a large role in speech initiation in the dominant hemisphere. The posterior SMA shows activation during the executive component of movement tasks [76]. Postoperatively, the SMA syndrome has been described as transient akinetic mutism (speech arrest) and hemiparesis (or apraxia) with potential long-term disorders involving rapid alternating movements of both hands [77]. Like the premotor cortex, the SMA is activated bilaterally during unilateral muscle movement with increasing complexity showing higher activation and can be activated during imagined complex movements [70].

19.4.2.4 Further Motor fMRI Activations of Interest

Further activity on fMRI can be seen in the anterior cingulate cortex (ACC) and the superior parietal lobule (SPL). The ACC shows enhanced activity on self-initiated movements compared to externally triggered movements [78] and is activated in preparation of motor response [79] yet is not motor-specific as, for example, activations in pain studies have been reported [80]. The ACC is generally believed to function in a wide range of cognitive control [81] including motivation and execution of goal-directed behavior. SPL activation is routinely noted in motor studies such as simple finger-thumb opposition [74], yet stronger activation is seen in sequential tasks [82] with a main postulated function in spatial planning of motor movements [83] and movement sequences [74]. However, unilateral lesions induce deficits not specific to the motor system, and include hemineglect and altered visual representations of space [84] as well as tactile apraxia (isolated disturbance of hand movements for use of and interaction with an object) [85].

19.5 Language fMRI

19.5.1 Introduction

Language fMRI presents a much greater challenge than motor fMRI. First, the activations of differing regions heavily depend on the task presented to the patient. Second, patients may not be able to cooperate due to differences in education, language, or deficits induced by the brain lesion itself. Moreover, it is difficult to assess patient cooperation on silent language tasks and boredom associated with free-thinking during control (resting) tasks. For this, we recommend interviewing patients after the examination to assess these variables. Again, closely supervised patient training before the examination is essential. Finally, individual languages can be represented on distinct cortical areas in multilingual patients [86], therefore the examination should be repeated for all languages of relevance for an individual patient. The fMRI language-tasks can be designed to investigate any language when patients are trained with a translating person; this is easily achieved when using pictures as triggers (Fig. 19.5).

Fig. 19.5
figure 5

BOLD activations of Broca’s area (left), the ventral premotor cortex (vPMC, middle), and Wernicke’s area (right)

Full size image

Preoperative language mapping on fMRI has two main tasks: determining hemispheric dominance (language lateralization) and localizing areas functionally important to language. As the brain activation associated with cognitive functions (such as language) is widely distributed and task-dependent, preoperative fMRI target activation of language areas (ROIs) may be reduced pragmatically to the classical model of language (Wernicke-Lichtheim), namely to the Broca and Wernicke language areas and the anatomical homologs of the right hemisphere (for language lateralization). However, such simplistic models do no longer correctly reflect the current understanding of cortical language functions. Typically, further language associated activation can be identified in the ventral premotor cortex, Exner’s, Dronker’s, and Geschwind’s language areas and distinct temporal areas (for further details see the following paragraphs). A combination of language fMRI with a DTI-tractography is recommended—at least of the arcuate fascicle. fMRI cannot currently replace the gold standard ECS in language localization, with a wide range of reported sensitivities (22–100%) [13, 14], a marked variability between patients [87], a dependence on tumor grade [10] and a concordance lower than motor mapping (67–100%) [9, 15, 16]. Language lateralization, on the other hand, has shown strong concordance with ECS reportedly on the order of 80 to 90% [17, 18]. One study reports fMRI to have an even greater accuracy in the prediction of postoperative outcomes compared to ECS for cases where the two tests were discordant [88]. Binder et al. conclude that fMRI is a valid alternative to preoperative Wada testing [17].

The following section is separated into cortical language areas visible on fMRI based contemporary understanding of language function known as the dual-stream model [89]. Language is no longer thought of in terms of the traditional Broca (expressive) Wernicke (receptive) model, rather in terms of cognitive function separated into phonetics (speech sounds), semantics (meaning of words), orthography (representation of written characters), and syntax (combining words into sentence structure) [16]. Driving this evolution was a multitude of factors: lesions to Wernicke’s area resulting in expressive language deficits [90], Broca’s area showing receptive function [91], and similar lesions between patients resulting in a variety of aphasia syndromes [92]. Furthermore, cortex neighboring the traditional Broca and Wernicke areas was shown to be more characteristic of the traditional area than the area itself! Corroborating these findings, Dronkers revealed the lesions in Broca’s original brain studies to span greater territories than his namesake area with parietal, insular, and subcortical lesions [93] included. Finally, this section does not describe the executive motor components of speech, as these areas are thoroughly discussed in the previous motor section.

Of note:

  1. 1.

    Recommended paradigms including slides for download can be found at www.asfnr.org, the website for the American Society of Functional Neuroradiology. Generally, paradigms activating primarily frontal/expressive regions include silent word generation, antonym generation, and object naming while sentence completion and passive story listening activate the temporal/receptive areas [94] (Fig. 19.6).

  2. 2.

    Language lateralization can be assessed visually or through the laterality index (LI) formula [95, 96]: LI: (Right hemisphere − Left hemisphere)/(Right hemisphere + Left hemisphere).

  3. 3.

    Neurosurgical priorities guiding resections based on ECS near language areas include identifying cortical areas producing speech arrest and anomia [97].

Fig. 19.6
figure 6

BOLD activations of the Broca (B), Wernicke (W), and Geschwind (G) areas in a patient with a perisylvian tumor. Left: word generation block design paradigm. Right: sentence generation block design paradigm

Full size image

19.5.2 Cortical Language Areas

19.5.2.1 Broca’s Area (Pars Opercularis and Posterior Pars Triangularis)

Paul Broca’s famous patients, Leborgne and Lelong, both presented with markedly reduced productive speech [98], attributed to lesions in the lateral left frontal lobe, resulting in the designation “Broca’s aphasia.” Broca’s area is now precisely defined in terms of the pars triangularis and pars opercularis of the inferior frontal gyrus [93]. While speech arrest is reported on ECS of the Pars triangularis [97] and the Pars opercularis [99] recent studies have shown a higher rate of speech arrest in an area known as the ventral Premotor cortex [99]. Furthermore, normal speech function has been reported following removal of Broca’s area [100, 101] but this is due to brain plasticity. For now, Broca’s appears to have a much wider range of cognitive function than initially thought, including semantic [102], lexical, phonological, and syntax [102, 103] in expressive as well as receptive speech [91] and higher order processing [104]. ECS results in a range of deficits including anomia as well as phonological and semantic paraphasias [99]. Rhyming tasks produce robust activation of Broca’s area [105]. For language lateralization producing robust activations of Broca’s and Wernicke’s area, word generation and silent word generation tasks are recommended [94, 106, 107] (Fig. 19.7).

Fig. 19.7
figure 7

BOLD activations of the Broca and Wernicke areas on axial (left) and sagittal (right) images with voxel numbers or lateralization purposes in a patient with atypical right hemispheric dominance. Paradigm: silent word generation

Full size image

19.5.2.2 Ventral Premotor Cortex (vPMC)

Located posterior and superior to the pars opercularis, the vPMC has been implicated in challenging Broca’s classical area for speech production, demonstrating 83% incidence of speech arrest compared with 35% for the precentral gyrus and 4% of pars opercularis on ECS [99]. The vPMC has many subcortical connections with the perisylvian network [108] and forms an important part of the final pathway for speech synthesis [109] and planification of articulation [73]. These observations, combined with its limited plastic potential [110] makes it essential to locate preoperatively. Tool selective activation was reported in a paradigm including tools, animals, faces, and houses with phase-scrambled images as control stimuli [111], even in the absence of subsequent activity. The specific hand movements associated with tools was postulated to stimulate regions of the brain (including the vPMC) housing information on the action associated with their use. Semantic tasks have also been shown to activate the vPMC [112]. While on the topic of the premotor cortex, a brief word of mention to Exner’s area is warranted, with a recent summary recommending its localization on preoperative language fMRI [91]. Exner’s area is located on the posterior aspect of the middle frontal gyrus with a reported function converging abstract orthographic representations into a planned motor sequence—isolated lesions were associated with deficits in the handwriting of words [113] while handwriting of numbers remained unaffected [114].

19.5.2.3 Dronkers’ Area (Precentral Gyrus of the Insula, Fig. 19.8)

In a lesion-based study, Dronkers observed that all 25 stroke patients with lesions to the precentral gyrus of the insula displayed speech apraxia, while 19 stroke patients without lesions to this area had no signs of such deficits [115]. Further studies corroborate the idea with the area being involved in the coordination of complex articulatory movements before speech is executed [116]. A metaanalysis of verbal fluency tasks using fMRI data incorporates the anterior insula as a node in a network responsible for the formulation of an articulatory plan, receiving word selection, and planned speech phonetics from the pars opercularis and relaying it to the cerebellum and caudate nucleus in a parallel fashion. From there, the modulated planned motor movement is integrated by the vPMC before being sent to M1 for execution [117]. Activation of the anterior insula can be seen, for example, in paradigms focusing on vocalized syllable complexity [118].

Fig. 19.8
figure 8

BOLD activations of the Broca (B), Wernicke (W), and Dronkers areas

Full size image

19.5.2.4 Wernicke’s Area

Wernicke’s area is located at the posterior aspect of the superior temporal gyrus (pSTG, exact definitions vary [119]) and has a vital role in processing units of speech sound known as phonemes. Originally described as the site of the brain responsible for language comprehension [120], current literature attributes speech production to the forefront of the pSTG’s function [90]. ECS, for example, has been shown to produce speech arrest [97, 99]. In fact, while areas around the classic Wernicke’s area seem critical to comprehension (middle temporal gyrus, anterior superior temporal gyrus, superior temporal sulcus, and angular gyrus [121, 122]), the classical Wernicke area itself is not directly implicated. These observations parallel those seen in the evolution of our understanding of Broca’s area: the neighboring cortex has become increasingly relevant and new functions have been attributed to the traditional area.

Specifically, the pSTG is responsible for retrieving phonemes [123] before speech is executed, with lesions resulting in an output disorder incorporating incorrectly chosen phonemes into words known as phonemic paraphasia [124, 125]. The cortex located directly anterior to Wernicke’s area, the medial superior temporal gyrus (mSTG), has implications in the comprehension of phonemes. The mSTG receives low-level auditory information from the primary auditory cortex (rostral Heschl’s gyrus [125, 126]) bilaterally—unilateral lesions are compensated by the contralateral hemisphere. Bilateral lesions, however, can cause pure word deafness [125]. Wernicke’s is robustly activated during sentence completion tasks [127].

19.5.2.5 Geschwind’s Area (Angular and Supramarginal Gyrus)

While Geschwind’s area is a combination of the angular and supramarginal gyrus, the two gyri have markedly different language functions: The angular gyrus is largely semantic [128] dealing with word meanings while the supramarginal gyrus is largely phonologic, dealing with word sounds [129]. The angular gyrus has a major role in reading and reading attainment [130] with lesions producing a range of cognitive deficits including alexia [131], agraphia [13] (simultaneous occurrence of these deficits is known as Gerstmann syndrome) anomia [132], sentence comprehension impairment [121], and a high rate of naming errors on ECS [133]. ECS of the supramarginal gyrus results in speech performance errors (apraxia) such as slurring or stuttering [133] as well as anomia [99]. Antonym generation tasks have been shown to activate these two areas [94].

19.5.2.6 Middle Temporal Gyrus

The middle temporal gyrus is part of the ventral semantic stream in the dual-stream model and is considered to be a major link between phonologic and semantic functions [134], integrating speech sounds from posterior temporal areas and relaying information to more anteriorly situated areas showing activation on higher, sentence-order levels [122]. Importantly, in a lesion based study, the MTG is one of only a few cortical areas shown critical to comprehension [121]. Activation on fMRI has been shown in phonetic studies such as speech versus nonspeech sounds [135] and phoneme discrimination [136] with variable activation on silent word generation tasks [94].

19.5.2.7 Temporal Pole

The temporal pole has special implications in surgery of intractable temporal lobe epilepsy. Here, removal of the dominant anterior temporal lobe can result in naming difficulties (more specifically people and place) [128, 137] with little or no naming deficits reported after resection of the non-dominant temporal pole [138]. For this reason, language lateralization in the ATL is warranted preoperatively.

The temporal pole plays a role in sentence-level semantic cognition [139] yet reliable activation on fMRI has proven challenging due to observations of semantic processing during resting states [140]. The proposed solution is to substitute the resting state control task with an arithmetic task, effectively interrupting the default mode activity in the ATL showing strong activation in the ATL across multiple centers [138].

19.5.2.8 Basal Temporal Language Area/Visual Word Form Area (BTLA/VWFA)

A discussion of naming deficits would not be complete without mentioning two areas on the fusiform gyrus known as the basal temporal language area (BTLA) and the visual word form area (VWFA). The BTLA is associated with a very high probability of lasting postoperative naming deficits [141], and while its exact function is not yet clear, suggestions of an important link between phonologic and semantic functions have arisen [134]. This is likely due to its proximity just anterior to the visual word form area (VWFA) housing representation of orthographic knowledge, showing activation during reading of printed words [142] when compared to pseudowords or speech [143]; damage to the VWFA results in pure alexia [125, 144]. The VWFA can be activated in sentence completion tasks based on reading [125, 145].

19.6 Optimization, Tips, and Pitfalls

19.6.1 Not Everything That Activates Is Essential

From the previous sections, it has become apparent that not all areas of activation on fMRI represent no-touch cortex on surgery. fMRI cannot differentiate essential (eloquent) cortex from participating “non-essential” cortex, which activates but is not essential for the task at hand. Language fMRI, for example, involves a range of processes including working memory and attention; these activations represent general processes. Differentiation of these cortices can only be done on ECS. fMRI, an activation-based study, differs fundamentally from ECS and Wada-testing, which are transient inhibitory studies. Furthermore, activations within the contrast-enhanced region of brain tumors should be considered artifacts until further research proves otherwise. The hemodynamic BOLD-response is prone to various tumor-induced alterations as compression of the vascular bed, neovascularization, and steal phenomena due to AV-shunting (details see below).

19.6.2 Conversely, Absent BOLD Activation Does Not Mean Inactive Cortex

fMRI is not an absolute study, it is a statistical, relative comparison of two functional states and many factors can lead to erroneous results. Furthermore, fMRI relies on assumptions such as intact neurovascular coupling. Medications such as acetazolamide [146] and ingested substances such as alcohol [147] decrease BOLD response by inducing vasodilatation while, as a side note, caffeine has actually been shown to boost the BOLD response [148]. The choice of paradigm also has an effect on fMRI results, in the preoperative context more importantly for language fMRI, as has been alluded to in the previous sections. While some factors can be optimized to increase the probability of a robust BOLD activation, some factors cannot be influenced. These include loss of vasodilatory response and impaired autoregulation in aging patients [149, 150], cerebrovascular disease with hypoperfused territories leading to ceiling effects [151], and susceptibility artifacts from hemorrhage causing signal drop out in postoperative and posttraumatic patients.

19.6.3 The Pathology Itself Can Nullify BOLD Activation

Finally, the pathology itself can render a study non-diagnostic. Tumors can decrease or even nullify BOLD signal through mass effect and edema, an effect shown to be dependent on distance to the tumor itself [152]. The neovascular structures induced by higher-grade tumors lose their ability to autoregulate and may already be maximally dilated [153] resulting in neurovascular uncoupling [154, 155] and therefore lower MR signal changes. While these reduced signal changes more pronounced in higher grade tumors [156], the altered pH levels, abnormal astrocytes, and disruptions in neurotransmitters in lower grade tumors also result in neurovascular uncoupling [157]. Here, mapping of neurovascular uncoupling is advisable using a breath-hold induced hypercapnic state; normal vessels will dilate while chronically dilated or pathologic vessels will fail to respond [157, 158].

AVMs can also result in impaired BOLD activation through associated flow abnormalities such as lower perfusion pressure [159] and adjacent steal effects with a demonstrated return of BOLD activation after embolization [160]. Moreover, AVMs have been shown to induce cortical reorganization to the ipsilateral [161] and contralateral [162] hemisphere that should not be mistaken for absent activation in the expected anatomical region.

19.7 Clinical Case

A 70-year-old male patient presented to the emergency room with dizziness. After neurological workup, MRI revealed a tumor in the right insular region with a small focus of contrast enhancement likely representing a high-grade glioma. The patient was right-handed but had a family history of right-hemispheric language dominance, thus preoperative fMRI was performed. Silent sentence generation and word generation paradigms confirmed atypical right-hemispheric dominance. Consequently, the neurosurgical approach was changed from that of a radical resection to a focused resection of the contrast-enhancing (presumably more malignant) area of the tumor. Postoperatively, no language deficits were found. Histology confirmed the diagnosis of Glioblastoma (Fig. 19.9).

Fig. 19.9
figure 9

Morphological axial FLAIR (a) and T1 contrast-enhanced (T1-ce) (e) images revealing a small area of contrast-enhancement within a larger, non-enhancing subinsular tumor in the right hemisphere. Functional BOLD-MR images overlaid onto sagittal T1-ce images reveal Broca’s (B) and Wernicke’s (W) areas using silent sentence generation (b) and word generation (f) paradigms. Location and lateralization of Broca’s and Wernicke’s areas on axial BOLD fMRI/T1-ce images with correlations of BOLD signal to modeled hemodynamic response function (HRF): left hemisphere image (c) (Broca’s area) and (g) (Wernicke’s area), right hemisphere image (d) (Broca’s homolog) and (h) (Wernicke’s homologue); graph beneath respective images

Full size image