Activation Methods

Magnetoencephalography

MEG is a noninvasive method of measuring brain activity by measuring the magnetic fields that accompany neuronal activity. Neural activity can be described as the generation and propagation of ion currents. The longitudinal current flow generated by several thousands of neurons firing synchronously can be detected at the scalp surface using a biomagnetometer. Unlike fMRI, which measures brain function indirectly by imaging cerebral vascular response to neuronal activity, MEG is based on electrical activity akin to EEG. Like EEG, MEG has excellent temporal resolution on the order of 1 ms. Unlike surface EEG, however, the MEG signal is not attenuated by the skull and scalp and has an excellent spatial resolution of approx 2 mm (21). MEG data are gathered using a biomagnetometer made up of wire induction coils arranged in an array covering the entire head. The magnetic fields produced by neural activity induce electric currents in these coils and can be used to reconstruct an image of the distribution of evoked neural electrical activity of brain function in real time. Magnetic source imaging is the coregistration of MEG data to a structural image to facilitate the anatomic-functional correlations and to incorporate this information into stereotactic neuronavigation systems (22).

MEG scanners require dedicated personnel as well as magnetic and radio-frequency-shielded rooms similar to those for conventional MRI. MEG is extremely vulnerable to environmental magnetic noise including the earth's magnetic field and the magnetically noisy environment in hospitals. Currently, MEG scanners are very expensive (>$2 million capital equipment costs) and have limited availability. Their use at this time is mainly restricted to centers pursuing research programs.

In neurosurgical practice MEG is used primarily in the presurgical evaluation of epilepsy patients to localize epileptogenic foci as well as functional areas that must be preserved during resection. For practical reasons, MEG scans are limited to interictal observations, but several studies have reported good correspondence between MEG recorded interictal spikes and seizure foci. Minassian et al. (23) reported, in a series of 11 children with neocortical epilepsy, a strong regional correspondence between the location of MEG-identified interictal spikes and ictal activity confirmed by subdural grid electrode recordings. Whe-less et al. (24) compared the accuracy of MEG for locating seizure activity with MRI, scalp video EEG, and interictal and ictal subdural grid electrode recording as determined by each method's ability to predict the clinical success of surgical resection. They found that MEG was second only to ictal grids and strips in predicting a positive surgical outcome but made no direct comparison between the anatomic location of seizure foci determined by each method (24). Mamelak et al. (25) compared MEG interictal data to intracranial electrode monitoring in 23 epilepsy patients. They found that MEG accurately localized seizure foci to the correct lobe and was therefore useful in guiding the placement of subdural electrodes, particularly in neocortical epilepsy (25).

MEG is also routinely used not only for identification of epileptogenic foci but also for functional mapping of sensorimotor cortex (26). MEG functional mapping of the auditory and visual cortex has also been performed. MEG task design is similar to the blocked design of fMRI experiments in that a stimulus is presented multiple times in alternating patterns of task and rest. The resulting evoked neuromagnetic signals are then averaged over several epochs to separate the signal produced by a focal population of active neurons from background activity. Identification of the central sulcus is most frequently achieved using passive sensory tasks such as electrical stimulation of the median nerve or tactile or vibratory stimulation of the hand and lower lip or tongue. Motor mapping is less common because MEG data acquisition requires smooth, well-controlled hand movements such as flexion and extension of one or more digits of the hand, which can be difficult for patients to perform.

The neurosurgical use of MEG has more recently expanded to stereotactic and image-guided surgery to aid in the safe resection of lesions threatening eloquent cortex (27,28). Several studies report good correlation between prop-erative MEG functional data and intraoperative maps of sensory- and motor-evoked potentials and electrocortical mapping (29-33). Several groups have merged functional MEG data with anatomic data in order to locate functional cortex near and within cortical lesions including arteriovenous malformations, gliomas, and brain metastases. Rezai et al. reported a technique of integrating MEG functional mapping data for both motor and sensory tasks into a stereo-tactic database for use intraoperatively as well as for preoperative planning. Their system combined MEG data with computed tomography (CT) scans,

MRIs, and digital angiography in an interactive stereotactic system and was used in 10 patients undergoing surgical resection of lesions involving the sensorimotor cortex. Similarly, McDonald et al. (34) report the successful combination of both fMRI and MEG data into a frameless stereotactic system that also incorporates digital registration of cortical stimulation sites. These techniques allow the simultaneous viewing of both structural and functional brain anatomy and their spatial relationship to brain lesions, which may allow the surgeon to resect more aggressively without violating functional cortical areas.

Positron Emission Tomography

PET was the primary functional imaging modality used in neuroscience research prior to fMRI and was the basis for the development of in vivo mapping studies of human functional cognitive anatomy. Positron emission tomography for functional mapping involves injecting a radioactive tracer, usually 150water, during the performance of both a behavioral task of interest and a control task. When brain areas are activated by a given task, there is a consequent increase in local cerebral blood flow. By comparing tracer uptake during the two conditions on a voxel-by-voxel basis, it is possible to make inferences about which regions are activated by performance of the task. PET may also be used to quantify resting or ictal metabolic activity; for these purposes a radioisotope with a longer half-life such as 18-fluorodeoxy glucose (FDG) is used.

The physiologic basis and methodology of PET are analogous to those of fMRI. In comparison with fMRI, however, PET has a relatively poor signal-to-noise ratio (SNR). That is, the signal of interest is only minimally distinguishable from background activity, or noise, generally requiring averaging of scans across multiple subjects or sessions in order to make a statistically significant observation about task-driven increases in metabolic activity for a given brain region. This low SNR makes PET a poor method for the presurgical mapping of an individual patient. Additionally, the spatial and temporal resolution of PET is only moderate. PET also requires the injection of radioactive tracers, making it an invasive procedure. This restricts the number and duration of PET scans on the same subject in order to avoid excessive radiation exposure and makes PET unsuitable in children and certain other populations.

Certain limitations intrinsic to PET constrain its use as a method to map complex task-related brain function. Since PET depends on the systemic distribution, half-life, and metabolic binding of its radioligand to its target pathway, PET is limited to examining those functions that can be sustained for several minutes, giving it very poor temporal resolution. The spatial resolution of PET is also quite limited owing to camera limitations and SNR-imposed limits on voxel size. Despite these limitations, PET has been used, and continues to be used, in many studies of brain function in both healthy subjects and patients. The extensive experience with PET as an activation test of brain function has also aided in the rapid development of fMRI as a technique owing to similarities in study design, analysis, statistical methodology, and interpretation (see fMRI section next).

PET has been used successfully in many brain mapping studies including investigation of somatosensory and motor function, vision, language (35), and memory and learning (36-38). Patient studies have investigated the effects of various pathological conditions (e.g., Alzheimer's disease [39-41]) as well as changes that accompany recovery of function (42-44). PET has been used to perform preoperative mapping of primary somatosensory motor cortex, language areas, and visual cortex (45). However, because of the aforementioned limitations, PET has generally been supplanted by fMRI in preoperative evaluation of eloquent cortex.

PET may also be used to examine metabolic abnormalities for the preopera-tive determination of seizure onset localization using FDG PET to quantify glucose metabolism (46,47). Interictally, regions of relative hypometabolism correlate highly with areas of epileptogenic tissue. The high sensitivity of PET is able to detect areas of hypometabolism that appear structurally normal on MRI and that may even demonstrate no histopathological abnormalities after resection. Ictal studies involve injecting the radioligand during a seizure, which may demonstrate hypermetabolism (although single-photon emission tomography [SPECT] is more commonly used in this setting owing to its relative simplicity). In the presurgical evaluation of candidates for epilepsy surgery, FDG PET data are combined with clinical information, scalp EEG, and structural MRI (1) to detect the appropriate side for anterior temporal lobectomy, (2) to select intracranial areas for microelectrode recording or grid placement if extracranial EEG provides insufficient localizing evidence, and (3) to establish the prognosis for seizure control following anterior temporal lobectomy (48,49).

Functional Magnetic Resonance Imaging fMRI is an emerging noninvasive brain mapping technique. It has been used extensively in cognitive science to study the brain basis of neurologic processes and is becoming more integrated into clinical practice as methodological issues are resolved and experience is accumulated across multiple centers. A major difference between fMRI (or PET) studies and IAT, intraoperative ECS, and TMS is that the former are tests of activation with certain tasks, whereas the latter are based on performance failure during brain inactivation. Inactivation tests are the standard for presurgical planning because they mimic the effects of surgical resection. Unfortunately, except for TMS, these techniques are highly invasive. Also, inactivation tests are not as amenable to the study of normal physiology or to task specificity as activation tests. Moreover, they do not allow detailed information to be available prior to surgery. Other issues include the inability to investigate the sulcal depths as well as underlying white matter tracts. fMRI has the potential to provide useful information on all these fronts non-invasively and preoperatively. Validation of the utility and accuracy of fMRI activation tasks in surgical planning is a major goal of fMRI research efforts.

Technique

Functional MRI localizes neural activity by measuring its correlate, regional cerebral blood flow. The most commonly used technique is blood oxygen level-dependent (BOLD) contrast imaging. When brain regions are activated during the performance of any activity, a neurally mediated vasodilation of capillaries and postcapillary venules occurs. This results in a relative increase in the ratio of oxygenated to deoxygenated hemoglobin owing to blood flow oversupply relative to increased neuronal utilization of oxygen (50). Because of the different magnetic properties of deoxyhemoglobin (paramagnetic) and oxyhemoglobin (diamagnetic), it is possible to measure these changes as alterations in the BOLD signal intensity on T2*-weighted images (51,52). Pixels whose signal intensity varies with the timing of stimulus presentation (with appropriate hemodynamic delay) represent activation by the task (53,54). Statistical inferences are made using correlation coefficients, statistical parametric mapping (SPM), or other methods to find those areas whose activity varies with the task paradigm. This information may then be overlaid on anatomic images forming functional maps. fMRI has spatial and temporal resolution far in excess of Wada or PET studies, particularly when using high field strengths (55). Moreover, fMRI is noninvasive and readily repeatable, as opposed to cortical stimulation testing. Patients can be studied sequentially, allowing the impact of surgery or other intervention to be assessed.

An alternative fMRI technique, known as perfusion imaging, detects changes in blood flow via a contrast (56) or spin-tagged bolus (57). Contrast-based techniques are relatively quick to perform, but require the injection of contrast material and are therefore not completely noninvasive. Spin-labeling techniques have the advantage of providing absolute blood flow values, which are being used for certain specific applications but are quite time consuming. Per-fusion-based techniques have the advantages of being both highly specific and less sensitive to motion artifact than BOLD. However, because of decreased sensitivity compared with BOLD, their use is limited to mapping areas with very strong intrinsic signal such as the motor cortex. Postprocessing for motion correction and statistics are similar to those of BOLD imaging. Perfusion imaging is affected by tumor vascularity and enhancement caused by blood-brain barrier breakdown. These effects are the basis for using these techniques to assess tumor histology, but the use of perfusion techniques for functional mapping in tumor patients may be confounded by these unknown effects (58).

Although many centers have all the necessary equipment to perform, analyze, and present fMRI studies, there are sufficiently complex issues of methodology to warrant caution, perspective, and rigorous quality control when implementing a new clinical fMRI program. fMRI can be performed on standard 1.5-T (or less) clinical MRI scanners. Higher field strengths such as the 3-T scanners currently being installed in some research and clinical facilities yield higher SNRs, allowing shorter scan times, more investigations, or better spatial resolution. The addition of faster gradients can allow faster acquisition times, thereby increasing signal or providing finer temporal resolution. As mentioned above, fMRI maps cortical function by measuring blood flow changes induced by neuronal activation in response to specific tasks. Thus the design and presentation of these task paradigms are vital to the success of the fMRI localization for a given function.

Currently, fMRI analysis is highly labor intensive, requiring personnel specially trained to administer the functional test paradigms, acquire the fMRI data, and perform data transfer and analysis. Functional MRI datasets are also very large, typically occupying up to a gigabyte or more of memory per study, so fMRI analysis also requires significant dedication of computer hardware and memory space. Turnkey software is being developed, but currently most data analysis is performed with a combination of freely available software (e.g., SPM [59]), commercial packages, and programs developed in-house. Finally, the interpretation of fMRI results requires an understanding of the fairly sophisticated statistical analyses required to process the data in addition to an understanding of the fMRI acquisition protocol, behavioral paradigms, and the patient's clinical status. Advantages

Functional MRI holds great promise for noninvasive mapping of the human brain in vivo. Unlike PET, fMRI is both noninvasive and readily repeatable. This can allow multiple investigations to map several brain functions fully, to investigate changes in brain activity under different conditions (e.g., medications) or to study the recovery process. Compared with PET, fMRI has a significantly higher SNR, allowing statistically powerful mapping of a single subject as required of a presurgical evaluation (60). fMRI data can also be combined with diffusion tensor imaging (DTI) of white matter tracts in and around lesions, allowing the visualization not only of the areas of cortical activation but also of the white matter tracts functionally connecting them to other areas (see the Diffusion Tensor Imaging section following for more information on DTI). In the future, fMRI may allow nonoperative and minimally invasive approaches to a variety of neurological problems by refining targeting of destructive lesions (e.g., radiosurgery, focused ultrasound, or radiofrequency ablation) and stimulation or neuromodulating procedures.

Challenges fMRI has its own limitations. As fMRI localizes neural activity by measuring its correlate, regional cerebral blood flow, findings may be affected by the many physiologic variables that influence cerebral perfusion or neurovascular coupling. Many of these variables remain to be studied, such as the effects of medications, changes occurring with normal aging, metabolic abnormalities, and the effect of mass lesions. Such limitations have so far limited the acceptance of fMRI in the clinical realm. Applying fMRI to neurosurgical problems also presents an opportunity to test directly the correlations between brain activity and the fMRI signal. Therefore, efforts to validate fMRI against the gold standard of intraoperative ECS are extremely important.

Surprisingly complex issues arise when one is performing, interpreting, and applying fMRI studies in a clinical setting (61). The complexity and resultant questions increase as one progresses from mapping relatively straightforward motor areas to less well understood cognitive areas. Outstanding questions and challenges include the following:

1. Interpretation of activations. There may be multiple areas outside the conventionally designated eloquent cortex that respond to a particular behavioral paradigm. It is unclear how to differentiate areas that participate in a task from areas that are required for the performance of this task. For example, whereas the cingulate gyrus has shown activation in a wide variety of cognitive paradigms including language production (62), resection of this structure does not result in a corresponding deficit. How to interpret this type of activation from a surgical planning standpoint may be unclear.

2. Validity. The precision of the spatial localization obtained by fMRI is still being studied and is probably affected by many patient variables. There are growing numbers of reports comparing fMRI and cortical mapping. However, these studies frequently define activations as much as 10 mm apart as "overlapping," even though this is a significant distance when one is trying to determine the limits of surgical resection. Krings et al. (63) sought to validate fMRI by comparing motor hand maps produced using four functional mapping techniques: PET, TMS, ECS, and fMRI. The fMRI maps strongly correlated with the maps produced by these other techniques; ECS-derived and fMRI maps fell within 1 cm of each other in 31 of 49 subjects and were neighboring (<2 cm) in 14 of 49. In the remaining patients, fMRI maps were uninterpretable because of low SNR or motion artifact (63). Several studies have examined the potential alterations in the somatotopic arrangement of the motor homunculus patients with tumors impinging on this area. fMRI reliably predicted cortical areas for hand and foot activation determined at the time of surgery by ECS from 82 to 92% of the time (64,65).

3. Reliability. Whether results obtained from fMRI studies are reproducible and reliable has significant implications for both its scientific and clinical applications. Whereas there is good evidence that on a group level activations associated with a given cognitive task are stable across institutions (66) and sessions (67), significant questions remain regarding the intrasubject reliability of given activations (68,69). A study of intersession differences concluded that a single fMRI study session provides only a snapshot in time of brain activation (68). As more data are acquired across several sessions, the certainty of activation of particular voxels increases. Another study indicated it may require approximately five sessions to achieve good statistical certainty (70).

4. Statistical thresholding. Functional maps derived from fMRI studies represent statistical probabilities that voxels are activated by a given task. This means, among other interpretation caveats, that the extent of activation will vary according to the statistical threshold is used. The significance threshold that corresponds in extent to the functional cortex found at operation by cortical mapping is dependent on the individual and cannot be predicted (71). When one is planning surgical resections immediately adjacent to eloquent cortex, the issue of how far to carry the resection depends critically on the extent of active cortex (Fig. 1; see Color Plate 1 following p. 112).

5. Influence of lesion type. Another important issue that merits further investigation is the effect of different types of lesions on the function maps generated by fMRI. It is reasonable to suppose that the functionality of adjacent brain will be different in infiltrative vs displacing lesions. Schreiber et al. (72) examined the alterations in fMRI maps for hand function in glial and nonglial space-occupying lesions. Motor cortex activation ipsilateral to nonglial lesions increased by 14%, in contrast to a 36% decrease with infiltrating gliomas. Their observations support the conclusion that gliomatous infiltration significantly alters cerebral hemodynamics and/or cor-

Fig. 1. Statistical maps of fMRI activation during a motor task in a patient with a low-grade lesion (yellow outline) involving the precentral area demonstrate the effect of varying the statistical threshold. As the threshold is made more stringent, the number of pixels exceeding the threshold diminishes, and the activation shown appears to shrink away from the tumor margin. (Courtesy of Drs. Lawrence Panych and Seung-Schik Yoo, Brigham and Women's Hospital Department of Radiology. See Color Plate 1 following p. 112.)

Fig. 1. Statistical maps of fMRI activation during a motor task in a patient with a low-grade lesion (yellow outline) involving the precentral area demonstrate the effect of varying the statistical threshold. As the threshold is made more stringent, the number of pixels exceeding the threshold diminishes, and the activation shown appears to shrink away from the tumor margin. (Courtesy of Drs. Lawrence Panych and Seung-Schik Yoo, Brigham and Women's Hospital Department of Radiology. See Color Plate 1 following p. 112.)

tical function, whereas nonglial lesions tend to displace but not destroy cortical neurons (72). Krings et al. (73) compared alterations in fMRI activation among tumor patients with varying degrees of hand weakness. They observed significant decreases in activation proportional to the severity of hand paresis but increases in the secondary motor areas remote from the tumor. They concluded that tumor tissue can interfere with cerebral hemodynamics and/or neural activity (73).

6. Selection of task paradigm. The choice and execution of a task paradigm as well as its associated "rest" condition will greatly affect the resultant activations. For simple motor mapping, true rest is probably an appropriate baseline task, but when exploring cognitive processes the choice of an appropriate baseline task becomes much more complex. In an auditory listening task to map language, for example, a resting baseline would result in more activation in areas such as the auditory cortex, which are outside of primary language areas. In comparison, the use of a highlevel baseline listening task such as listening to nonlinguistic stimuli would result in less activations in areas participating in nonlanguage auditory processing. Another critically important determinant of success in any fMRI study of patients is that they be asked to perform tasks that they are capable of performing (74). E.g., if a patient has a hemiparesis that interferes with his or her ability to perform a finger-tapping task, then any results from the study may be invalid.

Behavioral Protocols

The BOLD signal change associated with task activation averages 2% increase from background activation or noise. Therefore, to produce statistically significant observations, one must typically acquire and compare several sets of images ranging from 100 to more than 300 scans of the region of interest. For preoperative mapping and patient studies, the most common task paradigms use a block design in which activation tasks alternate with rest periods. This series is repeated through several cycles for an acquisition time of 3-10 min. A functional run may be repeated to increase SNR, or other functions may be investigated during typical fMRI scanning protocols, which last up to an hour. Block designs are generally used because they have excellent SNR, are easy for patients who may have cognitive or motor abnormalities to understand and to perform, and are fairly unaffected by changes in the hemodynamic response to neural activation (75). In addition, these types of studies are fairly straightforward to analyze, in contrast to some of the complex single-trial techniques gaining acceptance in cognitive science research.

These parameters limit the types of tasks that can be used for fMRI analysis to simple tasks that are discrete and repeatable. Most subjects are unable to tolerate sustaining the task/rest paradigm for extended periods, so fMRI scanning sessions tend to be limited to studying only one or two discrete functions. Because fMRI is highly sensitive to motion artifact that can irreparably corrupt data acquisition, tasks must be limited to activity that does not produce significant head movement. Most simple motor tasks below the shoulders do not interfere with BOLD acquisition (but see ref. 76, which suggests that there is measurable head movement with motor tasks and that it may be worse in patient populations). Any tasks associated with movements of the face and tongue can produce corrupting movement and artifact. This restriction on movement limits the usefulness of fMRI for speech testing to subvocal language generation, which does not allow for monitoring of patient performance and also may not engage all the cortical areas involved in overt speech production. It also complicates the comparison of fMRI studies with ECS for the purposes of validating fMRI in language studies, as ECS mapping necessarily uses speech performance and speech arrest as end points. However, several groups are developing techniques that overcome the difficulties associated with the use of overt speech generation (77,78).

Current Applications

Motor and Somatosensory: A few studies in small numbers of patients have demonstrated good concordance between fMRI localization of primary motor and somatosensory cortices and cortical stimulation (63,79-85). Krings et al. (63) compared motor activation maps produced by various functional imaging techniques (fMRI, PET, TMS, and ECS) in an effort to describe the relative accuracy of each method. They found that out of 50 patients, only 1 had fMRI-localized areas of activation, which contradicted the areas of activation demonstrated using these other techniques. In a separate study, the same group described difficulties with fMRI functional mapping encountered during this group's extensive experience (194 patients over 3 yr) with patient mapping (86). They report that fMRI mapping identified the motor regions in 85% of all investigated paradigms and only 11% failed to show fMRI activation. They further report that motion artifact was the most common reason for fMRI failure and that motor tasks involving the proximal limbs were more likely to produce such artifact than were tasks involving distal muscles. Interestingly, they also found that motion artifact correlated with the degree of paresis, implying that those patients who are most severely affected neurologically might be more difficult to map successfully.

Pujol et al. (87) report similar observations: they found selective and reproducible fMRI patterns of activation in 82% of 42 patients with space-occupying lesions. Of the 22 patients who proceeded to surgery, a 100% correspondence was found between the fMRI areas of activation and those identified by ECS. They failed to identify the central sulcus in 18% of these cases owing to damage to the primary sensorimotor region caused by the patients' lesions; this was correlated with the severity of hand paresis.

Atlas et al. (82) examined areas of activation surrounding gliomatous lesions compared with the contralateral side in such a way as to be able to demonstrate alterations in fMRI patterns of activation produced by tumor infiltration. They found displacement and decreased volumes of activation compared with the unaffected hemisphere. In an attempt to develop and validate improved methods of fMRI mapping in patients whose activation maps might be altered by brain pathology, Hirsch et al. (84) developed and tested an integrated battery of fMRI tasks in 63 healthy control subjects and 125 tumor patients. Sensitivity, as defined by the ability of specific tasks to activate the target regions, was 100% for motor cortex in control subjects. However, for patients harboring tumors, these figures were somewhat lower. As would be expected, both patients and controls demonstrated higher sensitivity in those cortical areas that are highly localized such as sensorimotor areas. Further work is needed to validate fMRI against the gold standard of cortical mapping in the more complex area of the supplementary motor cortex and to determine whether fMRI may provide additional information not available from cortical stimulation. Vision

Functional MRI has been used extensively in neuroscience studies to map primary and secondary visual areas. Primary visual areas can be robustly activated by photic stimulation (88), flashing checkerboard patterns, or other simple or complex visual stimuli. Preoperative mapping, although not commonly employed in this area, can provide a clear map of the retinotopic architecture of the striate cortex. Perhaps even more useful will be the development of DTI localization of the optic radiations in the temporal and parietal lobes. Language

The ability to define language areas accurately preoperatively is an important determinant of surgery for various pathological processes in both the frontal and temporal lobes. fMRI has been used both to lateralize (89-96) and to localize (82,84,97-101) eloquent language areas in the brain, to compare these techniques with the IAT, and to awake intraoperative cortical mapping.

Lateralization studies thus far have found excellent, although not constant, agreement between the two modalities (Fig. 2; see Color Plate 2 following p. 112). The most complex and difficult to interpret cases are those in which there is incomplete dominance with bilateral representation. To date, there is no consensus on which types of language paradigms (e.g., silent word generation, semantic tasks, reading, auditory comprehension) and analysis methods (e.g., lateralization

Was this article helpful?

0 0
Unraveling Alzheimers Disease

Unraveling Alzheimers Disease

I leave absolutely nothing out! Everything that I learned about Alzheimer’s I share with you. This is the most comprehensive report on Alzheimer’s you will ever read. No stone is left unturned in this comprehensive report.

Get My Free Ebook


Post a comment