by: G. Rees Cosgrove, M.D., F.R.C.S.(C), Bradley R. Buchbinder M.D., Hong Jiang, Ph.D.
Departments of Neurosurgery and Radiology, Massachusetts General Hospital,
Harvard Medical School, Boston, Massachusetts

Address correspondence to:
Address for Correspondence:
Emad N. Eskandar, M.D.
Massachusetts General Hospital
15 Parkman St. ACC # 331
Boston, MA 02114

E-mail: eeskandar@partners.org
Patient Appointments: 617.724.6590
FAX: 617.724.0339

Historically, neurosurgeons have mapped cortical function invasively by direct electrical stimulation of the cortex, either intra-operatively using a hand held stimulator or extra-operatively using chronically implanted subdural grids. These methods have proven value but obvious technical limitations. Classical cortical mapping requires a craniotomy under local or light general anesthesia with its associated discomfort, risk and expense. Only limited areas can be tested and fully 2/3 of the grey matter or the exposed surface is found along the depths of the sulci and is inaccessible to stimulation. Most importantly, the information cannot be used preoperatively for risk assessment, therapeutic decision-making and surgical planning.

Non invasive cerebral mapping techniques have thus evolved to localize functionally important cortical areas. They attempt to resolve the spatial and temporal distribution of neuronal activity during behavioral and cognitive tasks by measuring one of two physiologic responses: electromagnetic response or hemodynamic/metabolic response. Magneto-encephalography (MEG) or electroencephalography (EEG) measure electromagnetic response and positron emission tomography (PET) measures the hemodynamic/metabolic response. Both can localize certain cortical functions non-invasively but require dedicated, expensive units that are not widely available. To facilitate anatomic localization and clinical use, these modalities are typically correlated with magnetic resonance imaging (MRI) and/or computerized tomography (CT) using some type of image co-registration and fusion algorithm.

Modern MRI scanners can provide exquisite anatomical detail of normal cortex as well as demonstrate pathology. Images are generally viewed in the traditional 2-D orthogonal views ( axial, sagittal, coronal) or 3-D representations with surface renderings. Recent advances have now made noninvasive functional mapping feasible with MRI in the clinical setting. 7,11 Functional MRI (fMRI) can combine detailed anatomical information with precise physiological information to create a structural and functional model of an individual's brain. This information can be used preoperatively for surgical planning and intra-operatively for guiding surgical resection strategies. The combination of anatomical and physiological information makes fMRI an ideal tool for intracranial navigation.

Principles of fMRI

Basic neurophysiologic studies over many years have shown that there is tight coupling between cerebral activity, blood flow and metabolism.14 Since neuronal activity is associated with changes in regional cerebral blood flow (rCBF), hemodynamic measurements can provide an indirect assessment of neuronal activity.15 This principle forms the basis of the hemodynamic/metabolic mapping techniques that have been employed for many years in activation PET studies with excellent results.2,4,9 PET measures cerebral blood flow using tracers containing the positron-emitting isotope O15. The basic technique is elegant and simple: (1) acquire CBF images during a control task (e.g. rest) and one or more activation states (e.g. moving the fingers of the left hand); then (2) subtract the control CBF images from the activated CBF images to reveal regions of increased rCBF associated with task specific neuronal activation ( i.e. activated rCBF = task rCBF - baseline rCBF).

In 1991, Belliveau and coworkers at the Massachusetts General Hospital first reported the sensitivity of MRI to functional brain activity.1 Using MR images obtained rapidly after intravenous bolus injection of a paramagnetic contrast agent, they were able to demonstrate signal change in the human visual cortex after activation with photic stimulation. This signal change was presumably due to the increased concentration of contrast agent secondary to increased rCBF and was called the "contrast bolus tracking" technique. In 1992, Ogawa et al. described a second fMRI technique that did not require injection of an exogenous contrast material.13 This technique takes advantage of the inherent paramagnetic qualities of deoxyhemoglobin for use as an endogenous contrast agent. The underlying concept is that blood flow and deoxyhemoglobin concentration in cerebral capillaries and veins affects the signal intensity with magnetic resonance imaging. These signal intensity changes can then be used to indirectly measure changes in regional cerebral blood flows secondary to neuronal activity. The term blood oxygenation level dependent (BOLD) technique was coined to distinguish this from the contrast bolus tracking technique.12 BOLD imaging is completely noninvasive and has subsequently become the fMR imaging method of choice.

The theoretical basis for BOLD imaging is thought to be due to net changes in deoxyhemoglobin concentration that accompany changes in regional CBF.

In resting brain, there is a close correlation between rCBF and regional cerebral metabolic rates for glucose(rCMRglc) and oxygen (rCMRO2).15 With physiologic activation, rCBF may increase by as much as 50% in response to increased neuronal activity, far in excess of oxygen metabolic demands. The physiologic mechanisms by which this increase occurs is not fully understood although various mediators including neurotransmitters such as serotonin, acetylcholine, neuroactive peptides and nitric oxide produced by cortical neurons have all been implicated.6 During transient neuronal activity however, it is suspected that the brain may utilize anaerobic metabolism. The increased rCBF with only slightly increased O2 extraction results in a overall increase in oxyhemoglobin concentration and a relative decrease in deoxyhemoglobin concentration in the capillary and venous beds of the activated cortex.5 The signal changes appear to come from the microvasculature, capillaries and venules < 10 microns at 4 T and from larger venules and veins at 1.5 T.10 The net decrease in paramagnetic deoxyhemoglobin concentration then causes increased signal intensity on T2*- weighted MR images. Since these signal changes are quite small (2-6%), successful imaging requires acquisitions with high signal to noise ratio and subtraction techniques to detect the signal changes between a resting state and the activated state over time. The biological time constant of these changes is in the order of 2-3 seconds.

Initial fMRI studies were carried out on high field strength magnets (2.0 - 4.0 T). With higher field strengths, the magnitude of the functional signal change is significantly increased resulting in enhanced signal to noise ratio and improved spatial resolution. With stronger shielded gradient coils, improved electronics and better surface coils, conventional 1.5 T MR scanners can be used to obtain fMR images 3 but optimal studies still require high field strength magnets or special echo-planar technology. Echo-planar technology implies "fast" imaging with image acquisition times on the order of 70 - 100 ms.16 With appropriate relaxation times, this means complete volumes can be imaged every 1.5 seconds. This rapid echo-planar imaging (EPI) has several advantages including increased temporal resolution, shorter imaging time, opportunity for more signal averaging and increased sample volume. Depending on the acquisition parameters, this allows for 13 - 20 slices of the brain to be imaged simultaneously with average voxel dimensions of 3 x 3 x 7 mm3. In all of our studies to date, we have used a GE 1.5 T Signa Scanner modified for EPI by advanced NMR Systems, Inc.

Functional MRI generates large volumes of data and therefore computer hardware requirements include adequate memory with additional data storage capacity. Image processing, registration and rendering requires high-end graphics workstations.

To obtain fMR images for clinical use requires special imaging acquisition sequences followed by detailed computer image analysis.

Image Acquisition

After careful immobilization of the patient in the scanner, a structural MR data set is created using two volumetric T1-weighted SPGR imaging sequences. The first sequence ( TR 25, TE 5, flip angle 30o, FOV 22, 256 x192, 1 NEX) takes multiple thin axial sections ( 1.2 mm x 124 slices) and is designed to render cortical gyral anatomy and related pathology. The second sequence ( TR 28, TE 13, flip angle 30o, FOV 22, 256 x192, flow compensated, 1 NEX) is obtained after gadolinium administration to visualize the cortical veins. These 3-D structural image sets serve as the template upon which the functional date is superimposed.

For functional data sets, T2*-weighted echo-planar images using the BOLD technique are acquired in a standard head coil using an asymmetric spin-echo pulse sequence (TR 20, TE 70, offset -25ms, FOV 20, 64 x64). Usually 8 - 13 slices (7mm thick, slice interval 0 - 2.5 mm) are acquired during alternating epochs (30 s duration) of the resting and activated state for a total duration of 2.5 - 4.0 minutes.

A variety of activation tasks can be carried out depending on the requirements of the clinical situation and the area of the cortex one wishes to map. Primary sensory tasks involve presentation of cutaneous, visual or auditory stimuli while combined sensorimotor tasks involve active repetitive movements of the appropriate body part (tongue, hand or foot @ 1 Hz). Language testing generally involves a verb generation task and memory testing a visual encoding paradigm.

Computerized Image Analysis

After image acquisition is complete, detailed image processing, co-registration and rendering is required. First, all image volumes within and between acquisitions are corrected for motion using an automated algorithm. Next, the 3-D volumetric T1-weighted SPGR imaging sequences are segmented to display soft tissue, bone and cortical surfaces. Regions of interest (ROIs) including pathological lesions and cortical veins are also segmented. Next, the functional imaging sequences are averaged and sites of activation identified by statistical analysis of the signal - time course on a voxel by voxel basis using the Kolmogorov-Smirnov test. Changes in the signal intensity by more than 2 - 3 standard deviations from baseline are considered significant and segmented out as a cortical activation (ROI).

Finally, the functional MR images are then co-registered, fused and volume rendered with the anatomic images using both surface and volume based matching techniques [ANALYZE, Mayo Clinic] 8. All image processing is carried out on computer workstations (Hewlett Packard 740 Series) and presented in a variety of formats including two dimensional orthogonal slices along the principal axes, volume rendered image or movie clip. Final image selections can be displayed either on the monitor or printed out as hard copy on a color printer.

Since March 1993, we have performed fMR imaging studies as part of the presurgical evaluation in 31 patients with lesions involving or adjacent to critical cortical regions. There were 17 males and 14 females ranging in age from 9 - 62 years. Lesional pathology included 18 tumors, 5 AVMs, 4 cavernous angiomas, 3 cortical atrophy and 1 cortical heterotopia. Functional MRI was successful in defining the cortical function of interest in nearly all patients and was confirmed at surgery with conventional intraoperative cortical mapping (bipolar stimulation, 50 Hz, biphasic square wave, 2 ms duration, 1-12 ma) in 26 patients. In 5 patients, surgery was deferred because of the perceived risk of disabling neurologic deficit that might result from surgical intervention. In those patients who underwent surgery, there was excellent correlation between the cortical gyral and venous anatomy observed at surgery with the anatomy predicted by fMRI. Sensorimotor tasks activated contralateral primary motor cortex in all patients and supplementary motor areas in the majority. Language tasks demonstrated more variable localization in Broca's area, Wernicke's area, premotor cortex, temporal neocortex and cerebellum. The selection of language paradigms was of utmost importance in determining activation areas. There was close correlation (3-5 mm) between the functional mapping by cortical stimulation and that predicted by fMRI. The fMRI studies accurately predicted the location and extent of all subcortical lesions not evident even on close visual inspection of the cortex. No patient suffered any unexpected deficit post-operatively.

Clinical Examples

Case 1: This 15 year old right handed boy had a 3 year history of intractable supplementary motor seizures with speech arrest and secondary generalization. Routine MRI demonstrated a small lesion in the left superior frontal gyrus. Video/EEG monitoring suggested left frontal onset and surgical resection of the lesion was planned. Preoperative fMRI demonstrated the lesion in the supplementary motor area anterior to primary motor cortex. (Figure 1) At surgery under light general anesthesia, electrocorticography was initially unrevealing and the lesion was not evident upon visual inspection or palpation of the cortex. Recognition of the cortical topography and venous anatomy localized the lesion and insertion of an intracerebral electrode demonstrated nearly continuous epileptiform discharges. Motor stimulation of the pre-Rolandic cortex caused flexion contraction of the thumb, fingers and wrist while stimulation of the presumptive supplementary motor area via the depth electrode caused tonic elevation of the right arm. Lesionectomy and minimal surrounding corticectomy was performed. Pathology demonstrated a dysembryoplastic neuroepithelial tumor. The patient made an uncomplicated recovery from surgery without motor deficit and has remained seizure free for one year.

In this case, the fMRI accurately identified the primary and supplementary motor areas but more importantly allowed for precise localization of the subcortical lesion by correlation of the exposed cortex with that predicted by the fMRI.

Case 2: This 39 year old right handed woman had a 10 month history of focal sensory seizures beginning in her left foot and then gradually ascending up her leg to her trunk. Routine MRI revealed a non-enhancing lesion in the right superior parietal lobule. FMRI was performed during hand and foot activation to demonstrate the relationship of the tumor to the primary Rolandic cortex. (Figure 2) Surgery was performed under local anesthesia with cortical mapping which confirmed the location of the central sulcus and the somatotopic distribution of sensorimotor function as predicted by fMRI. The lesion involved the post-central gyrus but complete resection of the tumor was accomplished. Pathology revealed oligodendroglioma. Patient made a good recovery from her surgery with the expected cortical sensory loss in the left leg but only temporary motor deficit in the foot.

This case demonstrates the ability of fMRI to localize sensorimotor function even in the presence of an infiltrating tumor and its ability to accurately define its subcortical extent.

Case 3: This 36 year old right handed women experienced several months of mild cognitive and memory impairment. A large left intraventricular meningioma was discovered on routine MRI. fMRI was carried out to localize language cortex in order to plan the optimal approach and select the safest area in which to make a cortical incision. (Figure 3) At surgery, a parasaggital high parietal incision was made avoiding presumed language cortex and the tumor was completely excised. Post-operatively, she had mild transient confusion and a temporary visual field cut but no dysphasia.

This case demonstrates the usefulness of fMRI to clearly define complex anatomical and functional relationships to allow the neurosurgeon to select the optimal approach to a difficult subcortical lesion in the dominant hemisphere.

Case 4: This 22 year old right-handed man had a 9 month history of focal seizures characterized by an unusual sensation in his throat followed by involuntary reflex coughing and speech arrest. After a generalized tonic-clonic seizure, a MRI revealed a small enhancing lesion in the left frontal opercular region. FMRI was performed to localize the lesion and its relationship to sensorimotor tongue and language cortex.(Figure 4) At surgery under local anesthesia, the primary sensorimotor tongue and anterior language cortex was confirmed by cortical stimulation and the lesion was localized by recognizing the unique gyral topography overlying the lesion. A small corticectomy and removal of a gangliocytoma was accomplished without motor or language deficits.

This case exemplifies the exquisite combination of structure and function that can be obtained with fMRI. It is important to note however, that only the anterior language cortex was activated because the language specific task was a word generation task rather than a comprehension task. Wernicke's area, although present just posterior to the lesion resection and confirmed by cortical mapping, showed no activation demonstrating the importance of careful task paradigm construction and interpretation.

As has been demonstrated, fMRI can be used non-invasively to localize cortical function with respect to a surgical target. In this way, a more accurate assessment of risk relative to the planned surgical procedure can be formulated. The 3-D display of structure and function also allows for more accurate preoperative planning of the surgical approach and optimal exposure. Determination of functional cortex allows for selection of the most appropriate and safe cortical incision and trajectory. In many cases, fMRI can lateralize and localize language function thereby determining hemispheric dominance. If initial success with paradigms for testing memory are validated then intra-carotid amobarbital tests may soon become unnecessary.

Intra-operatively, regions of interest can be superimposed on the patients soft tissue/scalp rendering and their locations with respect to normal anatomical landmarks (i.e. eyes, glabella, occiput, ears) can be seen, measured and used to plan scalp incisions and bone removal. Once the dura is opened, the fMR images display gyral anatomy and superficial veins which can direct cortical incisions and define appropriate resection margins. Visualization of these cortical veins is very important and they prove to be valuable stereotactic landmarks. As the brain shifts after withdrawal of CSF, following administration of mannitol or during resection of tumor, the veins will shift along with the cerebral structures and the relative anatomy of the vein to the functional areas or lesion remains intact. fMRI provides even more information then visual inspection of the cortex because it can display the subcortical location of the lesion and its extent with respect to the cortical surface.

One of the most interesting applications that has yet to be fully investigated is the potential that fMR might be able to predict recovery from a variety of neurological events. These would include recovery after a stroke, head injury, tumor surgery, or other neurosurgical interventions. The ability to follow cortical function over time and its possible reappearance may provide a unique opportunity to observe reparative CNS processes or neuroregeneration.

A final application will be in the area of medical education where students can employ "electronic brain cutting" to study CNS anatomy and function. The ability to cut sections in any direction through the volumetric data set and to remove specific structures would be a valuable interactive teaching aid to learning basic neuroanatomy. The neurosurgical resident could attempt, using a variety of computer tools, to perform an operation electronically just prior to a planned procedure. This would prepare him/her for the actual surgery as well as explore optimal neurosurgical approaches for a variety of pathological lesions.

Discussion

Navigation is the science of determining the exact position of an object in space and then charting a course for guiding that object safely and expeditiously from one point to another. The ideal intracranial navigation system must perform in a similar fashion. It must accurately display cerebral structures and function that can be easily integrated and utilized intra-operatively by the surgeon. Neurosurgeons like navigators, not only must have a thorough knowledge of the science but also considerable experience and judgment.

Functional MRI is a new and powerful neuroimaging technique that can create an anatomical and functional model of an individual patient's brain. The concurrent 3-D rendering of cerebral topography, cortical veins and related pathology gives an unprecedented display of critical relational anatomy. Since stereotaxy means simply the three dimensional arrangement of objects, then fMRI may be the ultimate stereotactic system. It allows us to see through the scalp and cortex into subcortical areas which are not visually apparent. It accurately predicts cortical gyral and venous anatomy as well as the subcortical location and extent of lesions. But most importantly, it is capable of mapping specific cortical functions to anatomical regions thereby combining form and function.

This information can be used preoperatively to assess whether the patient is a candidate for surgery or whether some form of alternative therapy i.e. radiotherapy or radiosurgery might be more appropriate. It is extremely useful for presurgical planning by demonstrating lesions and eloquent cortex in relationship to a soft tissue rendering of the patient's head. Intra-operatively, this allows for optimal placement of the scalp and bone flaps. After exposure of the cortex, an optimal approach to the lesion can be selected based on the representation of the subcortical lesion with respect to critical functional areas. fMRI demonstrates relational anatomy and any intraoperative shift will affect all anatomical structures in the field therefore minimizing this source of localization error that can occur with frameless stereotactic systems.

Although direct visual inspection of the computer rendered image is all that is generally necessary in making intraoperative decisions, the integration of these data sets with a frameless stereotactic device may be useful in the future. In most instances however, displaying exactly what the surgeon will see at surgery obviates the need for an electronic pointing device. When one can identify the gyrus and adjacent vein as the exact area of cortex overlying a deeper lesion, there is little need of a frameless stereotactic localizing system to point at what your eyes can clearly see.

fMR has many advantages over other non-invasive methods of mapping cortical function. In comparison to activation PET studies, fMRI is widely available and does not use any radioactive isotopes. PET facilities are limited and have poor temporal resolution. In order to provide appropriate structural and anatomical information, PET studies must be co-registered with MRI or CT. fMRI will be widely available, is non invasive and easily repeatable. Localization is possible even in the presence of local pathology including arteriovenous malformations and cavernous angiomas. FMRI has no ionizing radiation and provides both structural and functional information with the same imaging modality.

Limitations

The major limitations of fMRI is that it is labor intensive and requires a dedicated group of physicians and imaging specialists.. Currently, the average functional MRI study requires 60-90 minutes for image acquisition and then 4-8 man hours for image registration, segmentation and rendering. It is therefore not appropriate for all patients and should be restricted to those patients with lesions involving or adjacent to eloquent cortex. Movement artifact can be a major problem and therefore patient cooperation and stable head fixation is imperative.

Automation of the image segmentation and processing will undoubtedly reduce the time and expense involved in creating these images. Methods for quantification of the signal changes during fMRI need to be developed so that one might be able to predict the outcome for various pathological conditions including stroke and head injury. Paradigms must be developed to investigate higher cognitive functions. The system should be integrated with frameless stereotactic devices and heads up display for maximal usage in the operating room.

Conclusions

Functional magnetic resonance imaging can accurately represent cerebral topography, cortical venous structures and underlying lesions. Functional activation appears to accurately localize appropriate cortical areas and these studies are feasible in the presence of local pathology. It is extremely useful in presurgical planning as well as assessment of operability. Intra-operatively, it shows a great promise in being able to define the exact location and extent of lesions with respect to surrounding functional cortex.

• (1) Belliveau JW, Kennedy DN, McKinstry RC et al: Functional mapping of the human visual cortex by magnetic resonance imaging. Science 254:716-719,1991.
• (2) Colebatch JG, Deiber M-P, Passingham RE et al: Regional cerebral blood flow during voluntary arm and hand movements in human subjects. J Neurophysiol 65(6):1392- 1401,1991.
• (3) Connelly A, Jackson GD, Frackowiak RS, et al: Functional mapping of human primary cortex with a clinical MR imaging system. Radiology 188:125-130,1993.
• (4) Fox PT, Burton H, Raichle M: Mapping human somatosensory cortex with positron emission tomography. J Neurosurg 67:34-43,1987.
• (5) Fox PT, Raichle ME: Focal physiologic uncou[KHORMC1]pling of cerebral blood flow and oxidative metabolism during somatosensory stimulkation in human subjects. Proc Natl Acad Sci 83:1140-1144,1986.
• (6) Iadecola C: Regulation of the cerebral microcirculation during neural activity: Is nitric oxide the missing link. Trends Neurosci 16:206-214,1993.
• (7) Jack CR, Thompson RM, Butts RK eet al: Sensory motor cortex: correlation of presurgical mapping with functional MR imaging and invasive cortical mapping. Radiology 190:85-92,1994.
• (8) Jiang H, Robb RA, Holton KS: A new approach to 3-D registration of multimodality medical images by surface matching. Visualization in Biomedical Computing SPIE 1808:196-213,1992.
• (9) Leblanc R, Meyer E: Functional PET scanning in the assessment of cerebral arteriovenous malformations. J Neurosurg 73:615-619,1990.
• (10) Menon RS, Ogawa S, Tank EW, et al: Tesla gradient recalled echo characteristics of photic stimulation induced signal changes in the human primary visual cortex. Magn Reson Med 12:380-386,1993.
• (11) Morris GL, Mueller WM, Yetkin FZ, et al: Functional magnetic resonance imaging in partial epilepsy. Epilepsia 35:1194-1198,1994.
• (12) Ogawa S, Lee LM, Kay AR et al: Brain magnetic resonance imaging with contrast dependent on blood oxygenation. Proc Natl Acad Sci 87:9868-9872,1990.
• (13) Ogawa S, Tank DW, Menon R, et al: Intrinsic signal changes accompanying sensory stimulation: functional brain mapping with magnetic resonance imaging. Proc Natl Acad Sci 89:5951-5955, 1992.
• (14) Raichle ME: Circulatory and metabolic correlates of brain function in normal humans. In: Mouncastle VB, Plum KF Geiger SR (eds): Handbook of Physiology-The Nervous System V. Bethesda, MD: American Physioloical Society, 643-674,1988.
• (15) Sokoloff L: Relationships among local functional activity, energy metabolism and blood flow in the central nervous system. Fed Proc 40:2311-2316,1981.
• (16) Stehling MK, Turner R, Mansfield P: Echo-planar imaging: magnetic resonance imaging in a fraction of a second. Science 254:43-50,1991.

Figure 1: Vertex view illustrating lesion (yellow) in superior frontal gyrus just anterior to the supplementary motor areas (red). Right (green) and left (blue) hand activation tasks define the central sulcus.

Figure 2: Vertex view of oligodendroglioma (yellow) arising in the right superior parietal lobule and invading the post-central gyrus. Right (red) and left (blue) hand activation tasks define the primary sensorimotor cortex. Note the appropriate homologous representation along the central sulcus.

Figure 3:

A) Lateral view of laarge left intaventricular meningioma (yellow) with location of the primary sensorimotor cortex defined by hand activation (red) and speech cortex defined by specific language tasks (green).

B) Vertex view with superimposed cortical veins demonstrating the area for optimal cortical incision to approach tumor while avoiding important sensorimotor (red) and language (green) cortex.

Figure 4:

A) Lateral view with soft tissue rendering to demonstrate relative location of lesion (yellow), tongue sensorimotor cortex (red) and Broca's area (green) on the scalp. This can be useful to plan incision and craniotomy for optimal exposure.

B) Lateral view with soft tissues removed to demonstrate cortical location of lesion (yellow) in the subcentral gyrus with respect to tongue sensorimotor cortex (red), Broca's area (green) and superficial veins.