Magnetic source imaging: Basic principles and applications in neuroradiology

Magnetic source imaging: Basic principles and applications in neuroradiology

E v i d e n c e of I n n o v a t i o n Steven E. Seltzer, MD, Editor Magnetic Source Imaging: Basic Principles and Applications in Neuroradiology Jef...

1MB Sizes 1 Downloads 141 Views

E v i d e n c e of I n n o v a t i o n Steven E. Seltzer, MD, Editor

Magnetic Source Imaging: Basic Principles and Applications in Neuroradiology Jeffrey David Lewine, PhD, and William W. Orrison, Jr., MD

agnetic source (MS) imaging is a relatively new imaging technique that is proving to be an important addition to the array of diagnostic procedures used to evaluate brain function in patients suffering from tumors, epilepsy, trauma, stroke, and other neurologic and psychiatric conditions [1-4]. By combining functional data obtained via magnetoencephalography (MEG) with structural data obtained via magnetic resonance (MR) imaging, MS imaging provides clinicians with a detailed, case-by-case picture of the mapping of brain function onto brain structure. Just as the flow of an electric current within a wire generates a surrounding magnetic field, current flow within brain cells generates a surrounding neuromagnetic field. MEG involves the measurement of millisecond-by-millisecond changes in the spatial pattern of the summated neuromagnetic field generated by parts of the brain. By using simplifying mathematical models, MEG data often can be used to localize (in three-dimensional [3D] space) the intracellular currents that most actively contribute to the various components of the magnetic signal. By integrating this information with structural MR imaging data, MS imaging provides for identification of those brain structures responsible for these currents. In this fashion, MS imaging provides a spatiotemporal picture of the workings of the brain on a scale of millimeters and milliseconds. MS imaging affords two major advantages over the recently developed technique of functional magnetic resonance (FMR) imaging and more traditional functional neuroimaging procedures such as single-photon


emission computed tomography (SPECT) and positron emission tomography (PET). First, MS imaging provides a direct measure of brain electrophysiology, whereas FMR imaging, PET, and SPECT measure changes in brain metabolism and hemodynamics that are secondary and tertiary to the neurophysiologic changes of real interest. This is of concern because the exact nature of the coupling between brain electrophysiology, local metabolism, and regional cerebral blood flow is only moderately well understood, and it remains unclear how this coupling varies from brain region to brain region and how it is affected by disease and pharmacologic agents [5]. The second major advantage of MS imaging is its real-time resolution of brain activity. Information processing by the brain occurs on a scale of milliseconds, and only electromagnetic techniques are capable of capturing this. Regardless of how fast images can be taken with techniques such as PET and FMR imaging, the hemodynamic changes being measured take several seconds to develop. There are two principal electromagnetic methods: electroencephalography (EEG) and MEG. EEG involves measurement of electrical potential differences that extracellular neuronal currents establish between points of the skull surface. MEG involves measurement of the magnetic fields produced by intracellular neuronal currents. Both methods provide direct, real-time measurements of brain electrophysiology. However, the differing electrical conductivity properties of the brain, cerebrospinal fluid, skull, and scalp result in significant

From the Magnetic Source Imaging Facility, Department of Radiology, Veterans Affairs Medical Center, University of New Mexico School of Medicine, Albuquerque, NM. Address reprint requests to W. W. Orrison, Jr., MD, Department of Radiology, University of New Mexico School of Medicine, 2100 Ridgecrest Dr., S.E., Albuquerque, NM 87108. Received April 22, 1994, and accepted for publication after revision September 15, 1994.

Acad Radio11995;2:436-440 © 1995, Association of University Radiologists


Vol. 2, No. 5, May 1995

distortion of the electrical potential pattern at the scalp surface (relative to that at the brain surface). As a consequence, the spatial resolving power of traditional EEG is compromised, and it is difficult, without detailed knowledge of the exact shape and electrical properties of the skull and scalp, to specify with subcentimeter precision what brain regions are responsible for generating specific EEG signals. These same factors compromise EEG-related evoked potential strategies and quantitative EEG methods. Electrical conductivity barriers also cause some perturbation in the magnetic field pattern, but these effects are relatively small and inconsequential for most MEG applications [1]. METHODOLOGY AND TECHNICAL CONSIDERATIONS

The magnetic signals measurable outside of the head mostly reflect postsynaptic, intracellular current flow within the apical dendrites of pyramidal cells oriented parallel to the skull surface. Several thousand neurons must be synchronously activated to produce a detectable signal, and even the largest neuromagnetic signals (those associated with epileptic spikes) are only a few picoTesla (10-12 T) in magnitude--more than one billion times smaller than the magnetic field generated by MR scanners that may be located only a few hundred feet away. Given the small size of neuromagnetic fields, it is a significant scientific accomplishment that the relevant technology is now available for routine clinical use. To accomplish the remarkable feat of isolating neuromagnetic signals from the background noise, biomagnetometers (the devices used to measure neuromagnetic signals) are typically operated within a magnetically shielded room made of high-permeability magnetic materials that divert external magnetic fields away from the sensor. The most common type of neuromagnetic sensor is an axial gradiometer that consists of two interconnected induction coils wound in opposition. Temporal changes in the magnetic flux passing through the coils induce changes in the electric current flowing through the detector that are proportional to the change in the spatial gradient of the magnetic field. By measuring the spatial gradient of the magnetic field (rather than the absolute magnitude of the magnetic field), the detector is rendered highly sensitive to nearby magnetic sources (e.g., neurons located immediately beneath the detector) and relatively insensitive to far-field noise sources (e.g., distant power lines and MR scanners). The coils themselves are made of niobium wire that is maintained in a superconducting state by immersing the coil




in a liquid helium bath that is contained in a cryogenically insulated fiberglass vessel known as a dewar. The superconductivity of the system is essential for measuring the very weak magnetic signals. It allows current to flow within the wires without a resistive loss of energy. The detection coil is inductively coupled to a superconducting quantum interference device (SQUID) that acts as a high-gain, low-noise, current-to-voltage converter that provides the output for the sensor system. In summary, neuronal currents generate weak neuromagnetic fields that induce currents within the detection coil; the currents in the detection coil induce currents in the SQUID ring; and the SQUID electronics generate a largeoutput signal proportional to the spatial gradient of the neuromagnetic field. These basic concepts in MEG measurement are illustrated in Figure 1. When there is a change in brain activity, the associated neuromagnetic field changes in both time and space. To determine where in the brain the actively changing neuronal populations are located, it is neces-

FIGURE 1. Basic principles of magnetoencephalography showing the relation between nerve ce[Is, neuromagnetic fields, and currents induced in the detection system. SQUID = superconducting quantum interference device.



sary to sample the spatial pattern of the magnetic field at each instant in time. In m o d e r n systems, this is accomplished by having multiple sensor coils positioned over the head. In the United States, the most prominent clinical system is the Magnes biomagnetometer of Biomagnetic Technologies (San Diego, CA). This system possesses 37 sensor coils arrayed over a spherical cap that samples the magnetic field pattern over an area of approximately 164 cm 2. Alternative commercial systems include those of Siemens (Islen, NJ), Neuromag (Helsinki, Finland), and CTF (Vancouver, British Columbia, Canada). The development of large-array biomagnetometer systems that allow clinicians to sample the magnetic field over large regions of the head with a single placement of the dewar has been instrumental in bringing MS imaging technology from the laboratory to the clinic, but it does not solve a fundamental limitation of biomagnetic methodologies. Whereas the external magnetic field pattern generated by a particular pattern of intracranial currents is unique, the inverse problem of specifying those intracranial currents that generated a particular set of extracranial measurements is ill-posed and without a unique solution, unless simplifying assumptions are used to make the problem mathematically tractable. Current data analysis strategies make two assumptions. The first is that the head can be modeled as a spherical volume conductor. This is a reasonable first-order approximation, and even in regions where the actual geometry of the head deviates most from sphericity (i.e., over the temporal lobes), use of the spherical assumption introduces source localization errors of less than 2 cm. The second assumption is more tenuous. The magnetic field pattern at each instant in time is modeled as though it were generated by a point-source current dipole intended as a mathematically equivalent representation of the actual current configuration. Provided that the recorded magnetic signal mostly reflects activation of a single focal cortical region (less than 2 cm3), this model is quite good and certain types of brain activity (e.g., that associated with movement of the digits of the hand) can be noninvasively localized with millimeter accuracy. VcSereas this single dipole-in-a-sphere model is by no means appropriate for characterizing all brain activity, its clinical utility for characterization of certain types of activity has been demonstrated repeatedly. The key to MS imaging formation of MS localization images is reconciliation of MEG and MR imaging coordinate frames. As part of the MEG experiment, a 3D digitizer is used to specify the positions of the sensor 438

Vol. 2, No. 5, May 1995

coils within a head-centered coordinate system defined by the nasion (located at the bridge of the nose) and the tragus of each ear. These points are readily identified on MR image sections, and straightforward procedures for axis rotation and translation allow MEG space to be m a p p e d into MR imaging space.


As outlined shortly, MS imaging is particularly useful in (1) the preoperative mapping of sensorimotor cortex in neurosurgical patients; (2) the characterization and localization of epileptiform activiW; and (3) characterization of the abnormal spontaneous rhythms that are prominent in a wide range of neurologic disorders. Neurosurgical intervention in patients with neoplasm s and vascular malformations is often complicated by simultaneous desires to spare healthy tissues and effect complete removal of pathologic regions. The sparing of sensorimotor cortex is particularly important because surgically induced hemiparesis markedly reduces the quality of postoperative life. Unfortunately, in cases of space-occupying masses in frontal or parietal regions, it is often difficult or impossible to use neuroanatomic imaging procedures (MR imaging and CT scanning) to specify the location of the lesion relative to the central sulcus (the divider between precentral motor and postcentral somatosensory cortices). The traditional means for dealing with this situation is to record somatosensory evoked potentials directly from the cortical surface at the time of operation or to electrically stimulate the brain during surgery (to localize motor regions). There are two main problems to this invasive strategy. First, many patients with parietal or frontal lesions never get to surgery because their lesions are incorrectly classified as inoperable on the basis of the spatial position of the lesion alone. Second, these intraoperative methods can be applied only after craniotomy and commitment to a particular surgical approach. MS imaging affords completely noninvasive, preoperative localization of sensorimotor cortex that provides better risk assessment and craniotomy site selection [6, 7]. The basic strategy in preoperative mapping examinations is to take advantage of signal averaging techniques. Stimuli designed to activate specific brain regions (e.g., tones for activation of auditory cortex, electrical pulses applied to the median nerve to activate somatosensory cortex) are presented multiple times, and data epochs spanning the stimulus event are signal averaged to isolate time-locked brain responses from spontaneous brain

Vol. 2, No. 5, May 1995


activity unrelated to processing the stimulus. For motor examinations, patients repeatedly move one of the digits of the hand while time-locked signal epochs are recorded. The resultant average event-related fields are believed to reflect activation along the processing pathway of interest, and components of the waveform are believed to reflect select activation of specific brain regions. Via dipole modeling of the magnetic field pattern associated with these components, the relevant functional brain regions can be localized. For motor, somatosensory, and auditory processing, intraoperative monitoring has repeatedly confirmed the millimeter accuracy of MS imaging for preoperatively localizing the relevant primary cortical areas [6-8]. The availability of temporal information is often critical in these types of examinations. For example, simple movement of a finger generates activity in multiple brain regions, including supplementary motor areas, premotor areas, primary motor cortex, and somatosensory cortex. In cases in which mass effects from lesions have distorted the local neuroanatomy, the time sequence of activation of brain areas provides the only clue as to which regions are motor and which are somatosensory. In several dozen cases, MS imaging data have led surgeons to alter their surgical approach to a lesion (to better avoid eloquent cortex), and in more than one dozen instances MS imaging has demonstrated that a lesion initially presumed to have been inoperable could be resected safely without induction of motor deficits (Fig. 2). The MEG functional mapping portion of presurgical

:.,, ,


+ ÷



MS imaging examinations takes only 30-60 rain to complete. The multislice volumetric MR imaging portion of the examination takes an additional 30 min, and in most instances MS localization images can be made available to surgeons within a few hours. One of the oldest uses of MEG is in the characterization of epileptic activity [9, 10]. The recent availability of large-array biomagnetometers, combined with the development of MS imaging algorithms for aligning MEG and MR imaging data, has brought this application to the clinical forefront. MS imaging is particularly useful in the localization of generators of interictal spikes. Although some spikes have complex spatiotemporal morphologies that are not well characterized by singledipole models, more complex spatiotemporal dipole modeling algorithms are now available for clinical use. These algorithms, in combination with the simpler single-dipole algorithms (which are appropriate for some spikes), can allow for precise localization of regions of epileptic pathophysiology (Fig. 3). MS imaging localization of epileptic pathology is as accurate as that afforded by alternative functional imaging methods (e.g., SPECT and PET), and it is proving to be especially useful in pediatric cases and patients with extratemporal lobe epilepsy, in w h o m traditional neuroimaging techniques often yield inconclusive or contradictory findings [11]. The normal MEG, like the normal EEG, is characterized by several spontaneous brain rhythms, the most prominent of which (alpha and beta) have frequencies

!~i.-..'::: ,,

._~_~ ..~ ..4~ - ~ -+- Jr-


+ ÷ ÷ ?Z2




sn i

s I








i i


i t

i i







FIGURE 2. Magnetic source imaging data from a patient with a left neoplasm. There is significant mass effect and edema associated with the lesion, and it was not possible to use neuroanatomic methods to identify the central sulcus. Tactile stimulation of the right index finger evoked a significant neuromagnetic response that was recorded with 37 detectors positioned over the left hemisphere (A). The response was characterized by a significant change in the magnetic field approximately 45 msec poststimulus (stimulus onset is marked by the vertical line in each trace). The magnetic field pattern at this time point (B) had a dipolar pattern characterized by a region of emerging flux (solid lines) and entering flux (dashed lines). Dipole modeling of the data revealed the source of the signal to be superior and anterior to the lesion (C).With this information in hand, identification of the central sulcus was made and subsequent surgery was directed away from the immediately anterior motor cortex. Intraoperative corticographic monitoring of the somatosensory evoked potential recorded at the brain surface confirmed the noninvasive magnetoencephalography inference on the location of the hand somatosensory representation to be accurate to within 3 mm. The neoplasm was resected successfully without the induction of neurologic deficits. 439



Vol. 2, No. 5, May 1995

assessing brain physiology perturbed by disease. In both the clinic and the cognitive neuroscience laboratory, MS imaging is providing new insights into the neural substrates of behavior. As larger sensor-array systems and more sophisticated data analysis algorithms become available, soon only a few minutes of examination time by this remarkable technology may be required to provide a complete physiologic assessment of brain function with millisecond and millimeter resolution. REFERENCES FIGURE 3. Example of magnetoencephalography (MEG) data recorded from a patient with epilepsy. The MEG waveforms show frequent interictal epileptic spikes. The adjacent magnetic source localization image shows this activity to originate from a small region of the left temporal lobe.

above 8 Hz. In cases of pathology, lower frequency rhythms (1-6 Hz delta and theta) often dominate the power spectrum. MEG appears to be more sensitive than EEG to abnormal low-frequency activity, and the combination of the MEG slowing index (the ratio of delta and theta power to alpha power) with dipole modeling strategies is proving to be highly effective in the identification and localization of pathology in a variety of conditions, including stroke, epilepsy, psychiatric dysfunction, and postconcussive syndromes. This is especially useful in the latter two conditions, for which other imaging techniques often are not sensitive enough to reveal subtle pathophysiology [12, 13]. CONCLUSION

In summary, MS imaging provides the best available combination of spatial and temporal resolution for


1. Lewine JD. Neuromagnetic techniques for the noninvasive analysis of brain function. In: Freeman SE, Fukushima E, Greene ER, eds., Noninvasive techniques in biology and medicine. San Francisco: San Francisco Press, 1990:33-74. 2. Sato S. Magnetoencephalography. AdvNeuro11990;54:1-284. 3. Hamalainen M, Hari R, Ilmoniemi RJ, Knuutila J, Lounasmaa OV. Magnetoencephalography: theory, instrumentation, and applications to noninvasive studies of the working brain. Rev Mod Physics 1993;65:413-498. 4. Lewine JD, Benzel EC, Baldwin NG, Orrison WW. Magnetoencephalography. In: Wilkins R, Rengachary S, eds. Neurosurge~ 2nd ed. New York: McGraw-Hill, in press. 5. Roland, P. Brain activation. New York: Wiley, 1993:11. 6. Gallen CC, Sobel DF, Waltz T, et al. Noninvasive presurgical neuromagnetic mapping of somatosensory cortex. Neurosurgery 1993;33:260-268. 7. Benzel EC, Lewine JD, Bucholz RC, et al. Magnetic source imaging: a review of the magnes system of Biomagnetic Technologies Incorporated. Neurosurgery 1993;33:252-259. 8. Lewine JD, Orrison WW, Maclin EL, et al. Event-related magnetic fields and neurosurgical practice. In: Advances in biomagnetism, 1993. Amsterdam: Elsevier, in press. 9. Barth DS, Sutherling W, Engel J, et al. Neuromagnetic localization of epileptiform spike activity in the human brain. Science 1982;218:891-894. 10. Rose DF, Smith PD, Sato S. Magnetoencephalography and epilepsy research. Science 1987;238:329-335. 11. Lewine JD, Orrison WW, Halliday A, et al. Magnetoencephalography: functional mapping in epilepsy. In: Jack CR, Cascino GD, eds. Neuroimaging in epilepsy. Boston: Butterworth, in press. 12. Lewine JD, Orrison WW, Astur RS, et al. Explorations of pathophysiological spontaneous activity by magnetic source imaging. In: Advances in biomagnetism, 1993. Amsterdam: Elsevier, 1994:xxx-xxx. 13. Lewine JD, Orrison WW, Sloan JH, et al. Neuromagnetic assessment of pathophysiological brain activity induced by minor head trauma. Neurology. Manuscript submitted for publication.