MAGNETIC SOURCE IMAGING: FUNDAMENTALS AND APPLICATION EXAMPLES

 

CONFERENCES
TOPIC: MISCELANEOUS

---

 
Abstract

Magnetic source imaging (MSI) is a noninvasive functional brain imaging technique which allows to investigate the relationship between brain structures and their function. This paper introduces the basic principles that make MSI possible, along with some concrete applications that demonstrate its usefulness as a valuable clinical tool for presurgical cortical mapping. Several functions of the brain can be studied with MSI, including spontaneous neurophysiological activity and responses to external stimulation of the auditory, somatosensory, and visual pathways. These applications, along with results from our ongoing studies demonstrating brain lateralization for language, suggest that MSI is a very promising tool in the area of basic physiology, clinical neurology, and cognitive neuroscience.
 

1 Introduction

Several medical conditions, such as brain tumors, arteriovenous malformations (AVMs), and medically intractable epilepsy, require surgical intervention. When planning surgery in patients suffering from such conditions neurosurgeons frequently face the dilemma of deciding whether resection of the affected tissue will cause greater harm than the pathology itself, especially when the cortical areas to be resected are involved in movement or in speech. In addition to proper detection and localization of the lesion, a surgical operation often requires identification of those few millimeters of an individualís brain which are responsible for sensation, movement, and language. This procedure, known as functional brain mapping, is crucial, because resection of these areas can have devastating results. Until recently, the most reliable means to map the human cortex has been the use of direct electrical stimulation of the exposed cortex [22, 29]. However, the limited number of recording electrodes that can be implanted and the necessity to place them near the lesion site have hampered the success of this procedure. Currently, however, the cortex can be mapped noninvasively through magnetoencephalography (MEG)?a method for estimating the sources of intracranial currents by measuring the magnetic fields outside the head. Unlike computed tomography (CT) and magnetic resonance imaging (MRI) which provide information on the brainís anatomy, MEG provides information regarding its function. Magnetic source imaging (MSI) is a procedure that combines neurophysiological data, derived via MEG, with neuroanatomical data, obtained via MRI or CT, to determine the relationship between brain structures and their function. It incorporates many of the advantages of other recently developed functional imaging techniques, such as positron emission tomography (PET) and functional MRI. However, its temporal resolution of one millisecond or less make it by far superior than all the other modalities, in that respect. In this paper we attempt to introduce the basic principles on which MEG is based and the necessary procedures that make MSI possible. Additionally, we provide a few concrete examples to demonstrate the usefulness of the MSI technique as a valuable clinical procedure, especially in the area of epilepsy management.

 
2.1 Neurophysiological signals

The human brain is composed of vast numbers of electrically active neurons and other supporting cells (e.g., glial cells) that are assembled in functional groups. In particular, the outer surface of the brain, the cerebral cortex, consists of a thin (average thickness 2.5 mm), highly compact (typical density 10⁵ cells/mm), intricate network (approximately10¹⁵ synapses) of cells arranged in layers [19]. During cell activation, large quantities of positive and negative ions? namely sodium (Na⁺), potassium (K⁺), and chloride (Cl^-_ )?cross the cell membrane, moving from the intracellular to the extracellular fluid, and vice versa. For all practical purposes, this ion movement is equivalent to a current flow, and it is responsible for all the externally recorded neurophysiological signals. In particular, the electrical potentials and the magnetic fields recorded on the scalp as the electroencephalogram (EEG) and the magnetoencephalogram (MEG), respectively, result mainly from the temporal and spatial summation of postsynaptic activity generated along the apical dendrites of pyramidalcells?a specific type of cortical neurons.

However, signals measured by the EEG primarily reflect extracellular activity in the brain, and they are subject to considerable attenuation and distortion as they pass through several layers of tissue of different conductivity to reach the recording sites on the scalp [7, 23]. For this reason, localization of EEG signal sources is relatively imprecise. MEG, on the other hand, detects the magnetic fields arising mostly from intracellular currents [23, 42]. Tissues in the scalp and skull are essentially "transparent" to these magnetic fields, thereby enabling MEG to achieve more precise localization of the true anatomic origins of spontaneous or induced neural activity [23].

2.2 Source identification

A detailed analysis on the relationship between the intracranial sources and the extracranially recorded signals is beyond the scope of this article, but excellent reviews can be found, for example, in [12, 42]. In general, neural activity can be mathematically represented as a primary source with current density j(r) in a closed volume G of finite conductivity s(r). Outside this volume, the conductivity and current density are zero. The potential v(r) inside the brain can be computed as the divergence of the source from Poissonís equation,

,

and, under quasistatic conditions, i.e., sufficiently small time derivatives, the associated electrical field E(r) is given by

The intracellular currents inside the closed volume G give rise to an electric field in the extracellular space which, in turn, results in currents that flow passively through the rest of the conducting medium. The secondary current density j_s(r) is given by Ohmís law,

.

The total current density j(r) inside the brain can thus be divided into a primary (intracellular) and secondary (extracellular) component, indicated as j_p(r) and j_s(r), respectively. Of particular clinical interest are solutions to the primary current sources only which, under some specific assumptions [23] that are fairly met in practical applications, can be computed as follows. The magnetic field b¥ (r) due to the primary current density jp(r) is given by the well-known Biot-Savart law,

where d = r ó r' is the distance between the observation point r r and the source point r', and m₀ is the permeability of free space which, in general, is assumed to be valid for biological tissues as well. Similarly, the potential v¥ (r) due to the primary current density jp(r) is given by

.

Determining the potentials and magnetic fields that result from the primary currents is known as the forward problem. Conversely, the inverse problem requires localization of the intracranial primary current sources that give rise to the externally measured magnetic field (MEG) and potential (EEG). The inverse problem has no unique solution, because of the infinite number of current distributions that satisfy the Biot-Savart law. To overcome this limitation, it is necessary to make assumptions regarding the location or the geometry of the source.
 

2.3 Mathematical modeling

Localization of the electrical sources underlying the measured magnetic fields (inverse problem) poses a very difficult mathematical problem, because there is a potentially infinite number of possible current distributions that satisfy the Biot-Savart law. To overcome the limitation of non uniqueness of the solution, it is necessary to make assumptions regarding the location or the geometry of the source.

The most common approach has been to describe the activated neurons in terms of the "single-dipole-in-a-sphere" model, a single equivalent current dipole (ECD) located in the center of a spherically symmetric homogeneous medium. Very often, both assumptions (i.e., sphericity and centrality) of this simplistic approach are met, and the resulting model is neurophysiologically plausible. Indeed, on one hand, the interior surface of the cranium sufficiently approximates a spherical and homogeneous medium and, on the other hand, the activity of a small cortical area can be approximated as a point source. The latter is true because pyramidal cell dendrites (the putative generators of activity) are arranged in a columnar fashion with an orientation locally normal to the cortical surface. Such an arrangement allows summation of the current fields generated by synchronous activation of a population of neurons within a small area of cortex. Therefore, well-localized activity can be adequately represented by an ECD located approximately at the center of this area having its moment normal to the cortical surface.

It is important to emphasize that the fields recorded on the scalp represent the activity of a large population of cells rather than the activity of a single neuron. Indeed, no single cell can produce a sufficiently strong field to be detectable outside the head. It has been estimated [42] that the moment Q of the equivalent current dipole corresponding to the activity of a single neuron is approximately 3´ 10^-13Am. Using the Biot-Savart law, one can compute the magnetic field b at a distance of about 4 cm from the neuron, which is a typical distance between cortex and sensor. The resulting value of b = 2´ 10^-17 T is beyond the capabilities of any instrument available today. Therefore, synchronous activation of a large number of neuron is required to produce the neuromagnetic fields measure outside the head. This number has been estimated [30, 42] to be close to 10⁴ or 10⁵.

Several powerful tools for the numerical solution of the above equations have been developed based on finite element techniques. It should be noted that these equations are valid for any source geometry, which may correspond to a single dipole, multiple dipoles, or a continuous distribution. However, the single dipole solution is the basic building block in all cases. Indeed, for numerical purposes, the solution to multiple dipoles can be derived from the single dipole directly from superposition, and a continuous distribution may be regarded as a collection of many dipoles distributed throughout the source region. Nevertheless, alternative approaches, such as multiple dipole [26, 36] and distributed source [1, 6, 20] models, have been reported in the literature holding the promise for more realistic representations of complex neuromagnetic data. However, the validity of these models has not been sufficiently tested, as yet.

2.4 Recording device

Neuromagnetic signals are many orders of magnitude weaker than the ambient magnetic noise, which is due to the earthís field and to the presence of ferromagnetic objects and electrical instrumentation. For example, the magnetic field typically recorded from the brain in response to somatosensory stimulation has a peak amplitude of approximately 100 fT [14, 23, 32], whereas in a hospital environment electromagnetic noise (power lines, elevators, MRI magnets, etc.) in extreme cases may be as high as 1 T (10¹⁵ fT) [23]. Therefore, to detect this kind of biological activity, it is necessary to use highly sensitive instrumentation and, at the same time, attempt to eliminate extraneous magnetic fields.

The principle of recording neuromagnetic signals is very simple: when a time-varying magnetic flux passes through an induction coil composed of one or more loops of wire, it induces a time-varying electrical current within the wire. In a typical coil, though, this current is quickly dissipated as heat by the electrical resistance of the wire. MEG measurements were practically impossible before the introduction of superconductive instrumentation. All biomagnetometers use special superconducting induction coils which have essentially no electrical resistance. Thus, even very small changes in magnetic flux will induce a certain amount of current within the coil. The induction coil is coupled to a second superconducting device, a so-called SQUID (superconducting quantum inter-ference device), which acts as a very low noise, ultra high gain, current-to-voltage converter. Typical biomagnetometers are composed of large arrays of pick-up coils each connected to a SQUID. The SQUIDs and induction coils are generally maintained in a superconduct-ing state by immersion within a liquid helium bath contained in a fiber glass superinsulated cryogenic vessel, known as dewar, at an operating temperature of 4.2^o K. Figure 1 shows a diagram of a SQUID array inside a multichannel MEG system.

Figure 1: Schematic diagram of a multisensor array inside a dewar. Each sensor, shown on the right, is composed of a detection coil connected to a SQUID.
 

Magnetic noise from non neuronal sources is a significant problem in MEG recordings. To reduce the amount of extraneous noise, the system is operated in a magnetically shielded room made of alternating layers of materials of high magnetic permeability and electrical conductivity (typically mu-metal and aluminum, respectively). The magnetic permeability of air is 1, whereas that of mu-metal exceeds 80,000. Consequently, external magnetic flux trying to penetrate the air-filled shielded room that contains the biomagnetometer is diverted away from the sensor by the high permeability walls of the room. A typical room reduces external magnetic noise levels to less than 10 fT/Ö Hz [11].

In addition to magnetic shielding, one can improve the quality of the MEG recordings by selecting an appropriate type of coil. Available designs today include single-loop magnetometers, which measure the magnetic field directly, and first-, second-, and third-order gradiometers, which consist of two or more loops that measure spatial changes of magnetic field. High-order characteristics can also be simulated via software from pairs of magnetometer and gradiometer hardware. In general, coils can follow an axial or planar geometry. The relative merits of different designs have been reviewed elsewhere [10]. Various examples of coils are shown in Fig. 2.

Figure 2: Various coil designs include planar magnetometers (a) and gradiometers (b) as well as axial magnetometers (c) and gradiometers (d). In the last case, a spatially uniform field B induces currents J and -J of opposite polarity which cancel out.

A gradiometer can be used to improve the signal-to-noise ratio, since it can selectively cancel out magnetic signals emanating from environmental sources located far from the source of neuromagnetic signals of interest. Such an arrangement is shown in Fig. 2(d) where the simplest configuration of an axial gradiometer is depicted: it consists of a detection (lower) and a compensation (upper) loop. A field B coming from an external noise source having a spatially uniform magnitude will thread through both loops of the gradiometer and will induce in each of them currents of equal magnitude J but of opposite polarity that will add to zero net current. The distance between these two loops (known as baseline) determines the achievable suppression of remote interfering signals: the smaller the baseline, the higher the suppression. Depending on the distance between the two coils, the field generated by a neural source close to the gradiometer will induce currents of different magnitude in the two coils, thus resulting in a nonzero net current in the gradiometer. On the other hand, any field generated far away from the gradiometer will result in zero net current, since the field magnitude will be practically uniform in space. A baseline of about 50 mm is optimal for magnetoencephalography [9] while a shorter baseline of, for example, 30 mm may suppress signals from deep sources in the brain.

In addition to suppressing environmental and instrumentation noise, there is also a need to attenuate extraneous magnetic signals originating from the patients themselves. In cases where patients have large amounts of dental work or other metal within their bodies, adequate demagnetization can usually be performed with a standard commercial tape eraser. Patients with implanted electromagnetically active medical devices?such as cardiac pacemakers, neu-rostimulators, and infusion pumps?generally cannot be studied using MSI because of the electrical interference from their devices. Also, equipment entering the room must be screened for large metallic components or electromagnetic activity that could introduce artifactual signals.

The cost and complexity of instrumentation led to the initial development of MEG systems containing only a few channels. These sensors were suspended from a mechanical gantry that allowed the dewar to be moved about a subjectís head, but any one placement had limited spatial coverage. More recently, however, large arrays comprising dozens of magnetometers and/or gradiometers have been developed (e.g., the 148-channel Magnes 2500WH, Biomagnetic Technologies, Inc., and the 306-channel Vectorview, Neuromag, Ltd.). These systems allow the simultaneous acquisition of neural signals over the entire head, thus eliminating the need for multiple recording sessions.

The need for cryogenics, however, has restricted the design of whole-head multichannel systems to rigid helmet shapes, which by necessity must be designed to accommodate the largest head expected. The coils are effectively further away from the scalp of subjects with smaller heads, thus reducing some of the sensor sensitivity.
 

The MSI procedure consists of several steps that culminate in the display of functional information onto high resolution anatomic images obtained by MRI (or CT). For this final step, it is necessary to translate MEG localizations into MRI coordinates.

The precise location of the measurement points on the scalp is determined electronically with reference to a Cartesian coordinate system anchored on three landmarks (fiducial points) on each subjectís head: two external ear canal points and the nasion. The line passing through the two preauricular points defines the y-axis of the system. The line perpendicular to the y-axis passing through the nasion defines the x-axis, and the line perpendicular to the x-y plane passing through the x-y origin defines the z-axis.

The MEG anatomic reference frame is established using a digital device known as sensor position indicator. A set of three receivers triangulate the signal from a stylus-type transmitter placed successively at several reference points on the subjectís head (typically the three fiducial points, the vertex, and the inion). Additionally, small vitamin E-containing capsules visible on MRI are placed on the fiducials. The locations of the common markers on the MR images and the MEG measurements serve for translating MEG locations into the MRI reference frame. Validation studies have shown that the accuracy of this approach can be as high as a few millimeters [11]. The same stylus transmitter is used to define the curvature of the head by tracing its surface. This allows the dipole fitting algorithm to define a "local sphere" to represent the head in the "single-dipole-in-a-sphere" model. Additionally, to minimize movement artifacts, the patientís head position is continuously monitored during the recording process using a motion detector.

After the MEG anatomic reference frame has been established, the fiducial points registered, and the patientís head digitized, the actual MEG recordings are obtained. The latter consist of time-varying magnetic field measurements at each detector position which, at a first glance, appear like the voltage fluctuations of the conventional EEG. However, unlike the conventional EEG in which the time-varying tracings are the final product, the MEG data undergo further analyses that allow the sources underlying the recorded activity to be localized.

The ECD model yields a description of the instant current dipole in terms of its location, strength, and orientation, along with an estimate of its reliability (confidence volume). Even though there is some variation across labs regarding acceptance criteria for the dipole solutions, a high correlation between data and model and a small confidence volume are always desirable. Moreover, some fits can be rejected for being physiologically unreasonable. For instance, dipoles of extremely high moment ((>200 nAm) are likely to be generated by movement artifacts. It is, therefore, possible to select a set of "best-fitting dipoles" to describe the data.

Once the best fitting dipoles have been identified, they are coregistered onto a complete set of axial, sagittal, and coronal MRI slices as described earlier. The resulting images are then printed on MRI films with different symbols and colors, each corresponding to a distinct type of MEG activity. For example, in our lab, the locations of interictal spikes are depicted with yellow triangles, whereas locations corresponding to somatosensory responses are depicted with green squares.
 

As mentioned earlier, the neuromagnetic signals of interest are largely due to primary currents from activation of specific brain areas in response to endogenous or exogenous events. However, because of intrinsic MEG properties, not all active primary current sources produce fields measurable outside the scalp; only the sources which are oriented tangentially to the skull have the best chance of being detected [23]. Dipolar sources oriented obliquely, radially, or markedly distal to the head surface have little or no chance of being detected. The latter is especially true, if second-or higher-order gradiometers, instead of magnetometers, are used by the sensor. In spite of this limitation, a variety of brain events of clinical interest have cortical sources of proper orientation close to the head surface and do lend themselves to measurement.

In general, two types of activity are of interest, spontaneous and elicited. A typical example of the former type is the recording of epileptogenic discharges, such as interictal spikes. The latter type of activity, typically known as evoked potentials in the case of EEG, results from external stimulation of a specific sensory pathway, such as the auditory, the somatosensory, or the visual tract.

In our earlier studies when limitations in technology allowed MEG systems with only a few channels?specifically, we used the seven-channel BTi model 607?the sensor had to be placed successively at several adjacent locations to cover the area of interest on the scalp. For example, to cover one hemisphere, the seven-channel system required placement of the sensor at eight different locations to obtain a total of 56 magnetic measurements out of which an acceptable field map could be constructed. However, in our current studies where we have been using a state-of-the art MEG system?namely, the 148-channel BTi model Magnes 2500 WH?the distribution of magnetic flux is obtained over the entire head simultaneously. Thus, the new generation systems eliminate the need for multiple recording sessions and, therefore, reduce the related problems of habituation and of changes in attention level which are known to affect evoked response components (see, e.g., [17]).

In the next sections we present in summary form several examples from studies undertaken in our laboratory to illustrate the process of measuring neuromagnetic activity and localizing the underlying sources.

4.1 Auditory Evoked Responses

The N1 is the most prominent evoked potential (EP) component occurring at a latency of approximately 100 msec after the presentation of transient auditory stimuli. A variety of stimuli can elicit this component and its magnetic counterpart N1m, the source of which can be easily identified and localized.

For instance, we have used computer-generated 1-kHz pure tones of 50 msec duration (10 msec rise/fall and 30 msec plateau) with an intensity of 100 dB normal hearing level (nHL) delivered to the subject monaurally through plastic tubes terminating in ear inserts (Nicolet TiP-300), while white masking noise of 60 dB nHL was channeled into the other ear (for details see [32, 33, 47]) .

Typically, one "epoch" of MEG signals consists of a few hundred milliseconds of activity prior to and following stimulation onset. To improve the signal-to-noise ratio of the evoked response, many similar epochs (typically 50ó 500) are collected and averaged, using the stimulus onset as a timing lock. An assumption underlying the averaging approach is that the neuronal responses to repeated stimulation are identical. In practice, however, some habituation of neuronal responsiveness is encountered, and this leads to diminished responses in the later acquired epochs. To reduce this effect, and to eliminate artifacts from periodic events, such as arterial pulsations or the 60-Hz power line artifact, the interstimulus interval is typically allowed to vary randomly within a predefined range. In this case, the stimulus delivery rate was varying between 0.33/sec and 0.5/sec around a mean of 0.4/sec. As always, subjects were tested in a magnetically shielded room (Vacuumschmelze, GmbH). Averaging of as few as fifty responses was enough to produce a clear N1m component.

The magnetic waveforms (averaged responses) resulting from auditory stimulation are characterized by peaks very similar to those of a typical EP (Fig. 3). The most prominent peak, the N1m which corresponds to the N1 of the EPs, occurs with an average latency of about 85 msec poststimulus.

 
 

Figure 3: Example of N1m magnetic (MEG) and electrical (EEG) evoked component resulting from auditory stimulation.

After measuring the magnetic flux distribution over the entire head, one can construct isofield maps, which typically contain two extrema of opposite polarity around the time corresponding to the peak of the N1m component. Figure 4 shows the instantaneous flux distribution at 85 msec poststimulus where the N1m component reaches peak amplitude.

Figure 4: Dipolar pattern of field distribution computed at the peak of the N1m component. Solid and dashed lines represent out-going and in-going flux, respectively.

The extrema represent points of maximum in-going and out-going flux, respectively. Such a configuration corre-sponds to flux generated by a dipole-like source. With the additional observation that the skull can, at least roughly, be approximated by a spherical conductor, one can see the justification for using the "single-dipole-in-a-sphere" model to analyze MEG activity. Typically, dipole localization algorithms determine the best-fitting local sphere, i.e., the sphere that best approximates the curvature of the actual surface of the head, as defined by several measurement points made at the beginning of each session and the particular channel group selected.

A rough idea about the location of the underlying source can be obtained from the isofield maps: if the distance between the extrema on the map is, say, D, then the source would be approximately located at a depth equal to Dp2 below the midpoint between the extrema. This observation was used in the older systems to guide placement of the sensor over the head, whereas in the newer systems it is used for selecting appropriate channel groups to be employed by the dipole-fitting algorithm.

As mentioned earlier, the dipole estimation algorithm attempts to solve the inverse problem by minimizing the difference between calculated (obtained by first solving the "direct problem") and measured fields. At the end of the procedure, the dipole parameters are considered estimates of the actual location, orientation, and strength of the source that gave rise to the recorded magnetic field. Since these estimates are not unique, it is always desirable to confirm independently the validity of the localization parameters. Confidence in the validity of the solution may increase if the sources are localized in cortical regions which are already known to generate the particular responses recorded. These regions can be visualized by projecting the estimated dipolar sources onto each subjectís MRI.

The estimated sources of the Nlm component in the aforementioned studies were consistently found to be on the floor of the Sylvian fissure, i.e., the primary auditory area of the cortex, as shown in Fig. 5. The reliability of the N1m localization estimates has been confirmed by several other investigators [13, 21, 39].

Figure 5: Localization of the N1m sources in the primary auditory cortex.

4.2 Somatosensory Evoked Responses

Localization of the central sulcus and the adjacent precentral and postcentral gyri is a fundamental objective of most MSI studies of patients with tumors located in the parietal or frontal cortex. Although mapping of the cortical representation of the motor system has been successfully accomplished with MSI [25, 40], the motor data to date are not as reproducible as the sensory [44].

The cortical extent of the somatosensory area can be approximated by obtaining MSI localizations corresponding to painless tactile stimulation of the digits, lips, and toes. Tactile stimulation is typically achieved by applying bursts of compressed air to a plastic diaphragm clipped to the subjectís fingertip, lip, or toe [2]. Based on empirical data, an air pressure of about 25 psi applied for a duration of 20ó50 msec has been found to be optimal. Detailed mapping of the sensory homunculus using an extension of this approach can be achieved [43]. In cases where tactile stimulation fails to evoke a cortical response, electrical stimulation can be used [15, 18] which, however, introduces an artifactual saturation of the SQUIDs that limits the detection of the early-latency responses.

In our laboratory, a typical protocol for somatosensory mapping consists of sequential stimulation of a patientís digits, lower lip, and toes, separately on the left and right sides. Approximately 500 single trials are required, each composed of 100 msec of prestimulus and 100 msec of poststimulus activity. The interstimulus interval is approximately 500 msec. The average responses obtained from each of the 148 channels are then digitally filtered (typically we use a second-order Butterworth bidirectional filter with a bandpass between 2ó40 Hz) to eliminate high frequency noise, low frequency artifacts, and baseline drifts.

After filtering, localization coordinates corresponding to each time point between 30 and 75 msec following stimulus onset are obtained using the ECD model. The x, y, z coordinates with best correlation between the measured and predicted fields are selected as the source location. In general, our acceptance criteria include a model correlation greater than 0.97, source strength (dipole moment) less than 200 nAm, confidence volume less than 1 cm 3 , and signal-to- noise ratio two-to-one. Figure 6 shows an example of the cortical areas activated after stimulation of a patientís left thumb, index, little finger, and lower lip.

Figure 6: Localization of sources resulting from somatosensory stimulation.

4.3 Visual Evoked Responses

Neuromagnetic responses elicited by visual stimulation have been used to study the retinotopic organization of the visual cortex in which different parts of the visual field are mapped to different cortical areas (see, e.g., [12]). Thus, in cases whereby a lesion involves the occipital cortex, visual evoked responses can be used preoperatively to assess potential postoperative complications. Figure 7 shows an example of the cortical areas activated after stimulation of a patientís left hemifield with a checkerboard pattern (check size 1.4^o, reversal rate 1/sec). The maximum responses were obtained in the right hemisphere, at a latency of approximately
130 msec.

Figure 7: Localization of sources resulting from left hemifield visual stimulation.

4.4 Epileptogenic Spike Source Localization

The most important data for localizing epileptogenic regions are ictal events. However, the large artifacts from muscle activity typically associated with seizures make localization of the underlying sources extremely difficult. Consequently, eventhough it is possible to record ictal activity, most studies of epilepsy report localizations of interictal activity only.

Typically, surface EEG recordings taken simultaneously with the MEG data help to identify epileptogenic activity as well as to recognize a variety of physiological and electrical artifacts. The EEG recordings are usually not used in the subsequent source localization analysis, although combined EEG and MEG source localization theoretically might provide more comprehensive data.

As an example, a few seconds of simultaneous MEG and EEG recordings are shown in Fig. 8 where, in addition to the artifacts resulting from the electrical activity of the heart, three interictal spikes can be seen.
 

Figure 8: Interictal spikes detected in simultaneous MEG (upper) and EEG (lower) recordings.

To localize the intracranial source of the activity, contour maps of magnetic flux are constructed and the sources of the activity are modeled as a series of ECDs computed at successive time points around the peak following the procedure explained earlier. At these time points, the activity most closely fits the single equivalent-current dipole model, and hence it can be reasonably localized to a small volume of brain tissue. The source localizations of all spikes obtained during the study of this patient are shown overlayed on MR scans in Fig. 9. In this case the MSI presentation demonstrates a focal area of interictal epileptic activity over the lateral surface of the left posterior temporal lobe.

Figure 9: Source localizations corresponding in part to the spikes shown in Figure 8.

4.5 Cortical activation during language tasks

Identifying brain regions that are important for language functions is often an integral part of the presurgical evaluation that epilepsy patients undergo. Such information is necessary for surgical planning in order to minimize the likelihood of postoperative deficits that could result from the resection of language-related areas that lie close to the epileptogenic region. At present, this task is accomplished through invasive techniques (e.g., [22, 29]). In a series of studies conducted in our laboratory we have systematically examined the reliability and validity of MSI as a noninvasive procedure for identifying brain areas that contribute to language function.

For this purpose, we have explored a number of language tasks including a semantic decision task, and two 13 continuous-recognition memory tasks for single words. In these studies, evoked responses are recorded for 1-2 sec after the onset of a word presented either visually or aurally. The intracranial generators of the magnetic responses are modeled as ECDs, and the number of ECDs obtained in each hemisphere during this period is used as an index of the extent of cerebral activation associated with the experimental task. Close examination of the location of ECDs found during late portions of the response (i.e., 200ó700 msec after stimulus onset) reveals a consistent distribution of generators both within as well as between subjects. A test-retest study with 12 subjects showed that the locations of the estimated sources are highly reproducible.

Additionally, the validity of these results is supported further by the following observations:
 

i) ECDs are localized in regions believed to play an important role in receptive language functions according to a wealth of evidence from functional imaging and lesion studies (e.g., [8, 16, 3, 34]). These areas include the posterior part of the superior temporal gyrus, and the supramarginal and angular gyri. Characteristic MRI slices with the localized sources superimposed are shown in Fig. 10.

Figure 10: Structures involved in language during engagement in an auditory (circles) or a visual (triangles) continuous-recognition memory task.

 

ii) MSI-based activity in these regions is not modality-specific: there is considerable regional overlap between the auditory and the visual version of the word recognition task.

iii) The activation pattern, as indexed by the number of ECDs in these areas, shows the expected left hemisphere-favoring asymmetry: in 39 out of 44 (91%) individuals the extent of cerebral activation in the left hemisphere is greater than the right hemisphere, across tasks.

iv) The importance of locations where ECDs are found for language has been verified through direct (intraoperative) electrocortical stimulation in more than 20 patients who have undergone both procedures (invasive and noninvasive MSI mapping) thus far.
 
 

The examples reported in the previous paragraphs give but a brief description of possible MSI applications for both research and clinical purposes. In particular, in the case of epilepsy, the usefulness MSI as a noninvasive tool to preoperatively assess the relationship of epileptiform activity with areas of eloquent cortex is immense. It has been estimated that as many as 300,000 patients with medically intractable epilepsy could benefit from surgery and that approximately 5,000 new surgical candidates appear each year in the United States [28]. Yet, less than an estimated 2,500 operations are performed annually mainly because of 1) limitations of the currently available presurgical evaluation procedures to successfully localize the epileptogenic focus, and 2) the necessity of the present surgical approaches 14ïto minimize postoperative deficits. Several studies suggest that MSI can provide important diagnostic information in assessing the risk of surgery and for optimal preoperative planning and, thus, it may lead to safer, faster, and more cost-effective clinical procedures.

Several other noninvasive neuroimaging techniques provide information about localization of cortical activity based on hemodynamic and metabolic responses to external stimulation or task engagement, such as PET and fMRI. However, in addition to poor temporal resolution (which is on the order of tens of seconds) and the necessity to administer some radioactive tracers, PET provides details about metabolic processes (e.g., glucose metabolism) that are believed to be correlated with local neural activity. Similarly, fMRI depicts changes in brain hemodynamics secondary to actual neuronal electrical processes. MSI, on the other hand, has two principal advantages over these techniques: 1) it provides a direct measure of brain electrophysiological activity (the recorded fields are mostly due to intracellular currents), and 2) it has a time resolution on the order of one millisecond or less.

Whole-head MEG brain mapping has recently received FDA approval for clinical applications and it is available only in a few centers around the world. As technology advances with the development of larger sensor arrays and general source modeling algorithms, and as more studies are carried out, the clinical relevance of this technique is be-coming more apparent. Future clinical applications of MEG mapping may include brain injury and stroke assessment, dementias, developmental disorders, as well as further characterization of higher cortical areas involved in attention, memory, and cognition.
 
 
 

This work was partially supported by NIH grant NS 29540-005A1. Data collection and analysis was performed at the MSI facilities of Hermann Hospital, Houston, TX.
 

[1] Baumgartner, C., Sutherling, W., Di, S., and Barth, D.S. Spatiotemporal modeling of cerebral evoked magnetic fields to median nerve stimulation. Electroenceph. clin. Neurophysiol., 79: 27ó35, 1991.

[2] Benzel, E.C., Lewine, J.D., Bucholz, R.D., and Orrison, W.W. Jr. Magnetic source imaging: a review of the Magnes system of biomagnetic technologies incorporated. Neurosurgery. 33: 252ó9, 1993.

[3] Binder, J.R., Frost, J.A., Hammeke, T.A., Cox, R.W., Rao, S.M., and Prieto, T. Human brain areas identified by functional magnetic resonance imaging. J Neurosci; 17: 353ó362, 1997.

[4] Breier, J.I., Simos, P.G., Papanicolaou, A.C., Zouridakis, G., Wilmore, L.J., Wheless, J.W., Constantinou, J.C., & Maggio, W.W. "Cerebral laterality assessment using magnetoencephalography: A comparison with the Intracarotid Amobarbital Procedure." Neurology, in press.

[5] Breier, J.I., Simos, P.G., Zouridakis, G., & Papanicolaou, A.C. "Temporal course of regional activation associated with phonological decoding." Journal of Clinical and Experimental Neuropsychology, in press.

[6] Crowley, C.W., Greenblatt, R.E., and Khalil, I. Minimum norm estimation of current distributions in realistic geometries. In: SJ Williamson, M. Hoke, M. Kotami, and G. Stoink (Eds.) Advances in Biomagnetism. New York: Plenum Press, 1989.

[7] Cuffin, B.N., Cohen, D. Comparison of the magnetoencephalogram and electroence-phalogram. Electroenceph. clin. Neurophysiol., 47: 132, 1979.

[8] Damasio, A.R., Damasio H The anatomic basis of pure alexia. Neurology 33: 1573-1583, 1983.

[9] Dossel, O., David, B., Fuchs, M., Kr¨ uger, J., Kullmann, W.H., and Ludeke, K.M. A modular approach to multichannel magnetometry. Clin. Phys. Physiol. Meas. 12 Supl. B: 75ó79, 1991.

[10] Flynn, E.R. Factors which affect spatial resolving power in large array biomagnetic sensors. Rev. Sci. Instrum.; 65: 922ó935, 1994.

[11] Gallen, C.C. Schwartz, B., Rieke, K., Pantev, C., Sobel, D., Hirschkoff, E., Bloom, F.E. Intrasubject reliability and validity of somatosensory source localization using a large array biomagnetometer. Electroencephal. clin. Neurophysiol. 90: 145ó156, 1994.

[12] Hamalainen, M., Hari, R., IImoniemi, R., Knuutila, J., Lounasmaa, O. Magnetoencephalography?theory, instrumentation, and applications to noninvasive studies of the working human brain. Rev. Mod. Physics, 65: 413ó497, 1993.

[13] Hari, R., Aittoniemi, M., Jarvinen, M.L., Katila, T. and Varpula, T. Auditory evoked transient and sustained magnetic fields of the human brain. Exp. Brain Res., 40: 237ó240, 1980.

[14] Hari, R., Pelizzone, M., M¨ akel¨ a, J.P., H¨ allstr ¨ om, J., Leinonen, L., Lounasmaa, O.V. Neuromagnetic responses of the human auditory cortex to on- and off-sets of noise bursts. Audiology; 26: 31ó43, 1987.

[15] Hari, R., Karhu, J., Hamalainen, M., Knuutila J., Salonen, O., Sams, M., Vilkman, V. Functional organization of the human first and second somatosensory cortices: a neuromagnetic study. Eur. J. Neurosci. 5: 724ó734, 1993.

[16] Hart, J., Gordon, B. Delineation of single-word semantic comprehension deficits in aphasia, with anatomical correlation. Ann. Neurol; 27: 226ó231, 1990.

[17] Hillyard, S.A. and Picton T. The N1 wave of the human electric and magnetic response to sound: a review and analysis of the component structure. Psychophysiology, 24: 375ó425, 1987.

[18] Kamada, K., Oshiro, O., Takeuchi, F. , Kuriki, S., Houkin, K., Iwasaki, Y., Abe, H. Identification of central sulcus by using somatosensory evoked magnetic fields and brain surface MR images: three dimensional projection analysis. J. Neurol. Sci. 116: 29ó33, 1993.

[19] Principles of neural science. E.R. Kandel, J.H. Schwartz, T.M. Jessell (Eds.), Third edition. New York: Elsevier, 1991.

[20] Kullman, W.H., Jandt, K.D., Rehm, K., Schlitt, H.A., Dallas, W.J., and Smith. W.E. A linear estimation approach to biomagnetic imaging. In: SJ Williamson, M Hoke, M Kotami, and G Stoink (Eds.) Advances in Biomagnetism. New York: Plenum Press, 1989.

[21] Kuriki, S., Okita, Y. and Hirata, Y. Source analysis of magnetic field responses from the human auditory cortex elicited by short speech sounds. Exp. Brain Res., 104: 144ó152, 1995.

[22] Lesser, R.P., L¨ uders, H., Morris, H.H., Dinner, D.S., Klem, G., Hahn, J., Harrison, M. Electrical stimulation of Wernickeís area interferes with comprehension. Neurology; 36: 658ó663, 1986.

[23] Lewine, J.D. Neuromagnetic techniques for the noninvasive analysis of brain function. In: Noninvasive techniques in Biology and Medicine. SE Freeman, E. Fukushima, and ER Greene (Eds.) San Francisco: San Francisco Press, 1990.

[24] Loring, D.W., Meador, K.J., Lee, G.P. Cerebral language lateralization: evidence from intracarotid amobarbital testing. Neuropsychologia 28: 831ó838, 1990.

[25] Morioka, T., Yamamoto, T., Mizushima, A., Tombimatsu, S., Shigeto, H., Hasuo, K., Nishio, S., Fujii, K., Fukui, M. Comparison of magnetoencephalography, functional MRI, and motor evoked potentials in the localization of the sensory-motor cortex. Neurol Res; 17: 361ó367, 1995.

[26] Mosher, J.C., Lewis, P.S., Leahy, R. Multiple dipole modeling and localization from spatio-temporal MEG data. IEEE Trans Biomed Engineer, 39: 541ó557, 1992.

[27] Nakasato, N., Toshihiro, K., Kanno, A., Ohtomo, S., Mizoi, K., Yoshimoto, T. Neuromagnetic evaluation of cortical auditory function in patients with temporal lobe tumors. J Neurosurg; 86: 610ó618, 1997.

[28] Surgery for Epilepsy. NIH Consens Statement Online; 8: 1ó20, 1990.

[29] Ojemann, G., Ojemann ,J., Lettich, E., Burger, M. Cortical language localization in left, dominant hemisphere: An electrical stimulation mapping investigation in 117 patients. J Neurosurg; 71: 316ó326, 1989.

[30] Okada Y. Neurogenesis of evoked magnetic fields. In: SJ Williamson, GL Romani, L. Kaufman, and I Modena (Eds.) Biomagnetism: An Interdisciplinary Approach. New York: Plenum Press, 1983.

[31] Pantev, C., Hoke, M., Lehnertz, K., Lutkenhoner, B., Fahrendorf G., Stober, U. Identification of sources of brain neuronal activity with high spatiotemporal resolution through combination of neuromagnetic source localization (NMSL) and magnetic resonance imaging (MRI). Electroenceph. clin. Neurophysiol., 75: 173ó184, 1990.

[32] Papanicolaou, AC, Baumann, SB, Rogers, RL, Saydjari, C, Amparo, EG, Eisenberg, HM Localization of auditory response sources using MEG and MRI. Arch Neurol; 47: 33ó37, 1990.

[33] Papanicolaou, AC, Tarkka IM Magnetoencephalography. In: Bigler ED (Ed.) Neuroimaging 1: Basic Science. New York, NY: Plenum Press; 1996.

[34] Pugh, KR, Shaywitz, BA, Constable, RT, Shaywitz, SE, Skudlarski, P, Fulbright, RK, Bronen, RA, Shankweiler, DP, Katz, L, Fletcher, JM, Gore, JC Cerebral organization of component processes in reading. Brain; 119: 1221ó1238, 1996.

[35] Ruohonen, JO, Ravazzani, P, Ilmoniemi, RJ., Galardi, G., Nilsson, J, Panizza, M, Amadio, S, Grandori, F, and Comi, G. Motor cortex mapping with combined MEG and magnetic stimulation. Electroencephal clin Neurophysiol Suppl; 46: 317ó322, 1996.

[36] Scherg, M. Buchner, H. (1993) Somatosensory evoked potentials and magnetic fields: Separation of multiple source activities. Psysiol Meas 14: A 35ó39.

[37] Papanicolaou A.C., Simos P.G., Breier J.I., Zouridakis G., Willmore L.J., Wheless J.W., Constantinou J.E.C., Maggio W.W., and Gormley W.B. "Magnetoencephalographic mapping of the language-specific cortex." Journal of Neurosurgery, 90: 85-93, 1999.

[38] Breier J.I., Simos P.G., Zouridakis G., and Papanicolaou A.C. "Relative timing of cortical activation during a word recognition task." Journal of Clinical and Experimental Neuropsychology, in press.

[39] Singh, K.D., Ioannides, A.A., Gray, N., Kober, H., Pongratz, H., Daun, A. Grummich, P., Vieth, J. Distributed current analyses of bi-hemispheric magnetic N1m responses to ipsi/contralateral monaural stimuli from a single subject. Electroenceph. clin. Neurophysiol., 92: 365ó368, 1994.

[40] Sobel, DF, Gallen, CC, Schwartz, BJ. Waltz, TA, Copeland ,B, Yamada, S, Hirschkoff, EC, and Bloom, FE. Locating the central sulcus: Comparison of MR anatomic and magnetoencephalographic functional methods. Am J Neuroradiol 1993; 14: 915ó925.

[41] Wada, J, Rasmussen, T. Intracarotid injection of sodium Amytal for the lateralization of cerebral speech dominance: experimental and clinical observations. J Neurosurg; 17: 266ó282, 1960.

[42] Williamson, S.J. and Kaufman, L. Analysis of neuromagnetic signals. In AS Gevins and A Remond (Eds.) Handbook of Electroencephalography and Clinical Neurophysiology: vol. 1, pp. 405ó448. New York: Elsevier, 1987.

[43] Yang, TT., Gallen, CC., Schwartz, BJ., and Bloom, FE . Noninvasive somatosensory homunculus mapping in humans by using a large-array biomagnetometer. Proc. Natl Acad. Scie USA, 90: 3098ó3102, 1993.

[44] Weinberg, H, Cheyne, D, and Crisp, D. Electroencephalographic and magnetoencephalographic studies of motor function. Adv. Neurol., 54: 193ó205, 1990.

[45] Zouridakis, G., Simos, P.G., Breier, J.I., and Papanicolaou, A.C., Functional hemispheric asymmetry assessment in a visual language task using MEG. Brain Topography, 11: 57-65, 1998.

[46] Zouridakis G., Simos P.G., Breier J.I., and Papanicolaou A.C. Assessment of Functional Hemispheric Asymmetries in a Visual Language Task via MEG. Abstracts of the Sixteenth Annual Conference on BiomedicalEngineering Research in Houston, p. 118, 1998b.

[47] Zouridakis, G., Simos, P.G., and Papanicolaou, A.C. Multiple bilaterally asymmetric cortical sources account for the auditory N1m component. Brain Topography, 10: 183ó189, 1998.

---