RESEARCH
Advanced electronics for the CTF MEG system
Development of the CTF MEG system has been advanced with the introduction of a computer processing cluster between the data acquisition electronics and the host computer. The advent of fast processors, memory, and network interfaces has made this innovation feasible for large data streams at high sampling rates. We have implemented tasks including anti-alias filter, sample rate decimation, higher gradient balancing, crosstalk correction, and optional filters with a cluster consisting of 4 dual Intel Xeon processors operating on up to 275 channel MEG systems at 12 kHz sample rate. The architecture is expandable with additional processors to implement advanced processing tasks which may include e.g., continuous head localization/motion correction, optional display filters, coherence calculations, or real time synthetic channels (via beamformer). We also describe an electronics configuration upgrade to provide operator console access to the peripheral interface features such as analog signal and trigger I/O. This allows remote location of the acoustically noisy electronics cabinet and fitting of the cabinet with doors for improved EMI shielding. Finally, we present the latest performance results available for the CTF 275 channel MEG system including an unshielded SEF (median nerve electrical stimulation) measurement enhanced by application of an adaptive beamformer technique (SAM) which allows recognition of the nominal 20-ms response in the unaveraged signal.
Magnetoencephalography: the art of finding a needle in a haystack
The brain's magnetic signals are much weaker than the magnetic disturbances inside the typical commercial magnetically-shielded room. Magnetic noise arises from far-field environmental sources (power lines, vehicles, etc.) and from near-field biological sources (electrically active tissues, such as muscle, heart, unwanted brain signals, etc.). Some form of inverse solution is generally used to solve for the sources that account for the MEG measurements. However, the inversion problem is non-unique and ill defined. Given the large amounts of noise and the non-uniqueness, how can MEG inversion succeed? One must provide methods for efficient attenuation of environmental noise, combined with MEG localization methods that are robust against the background clutter. Noise cancellation methods will be reviewed, and it will be shown that a combination of synthetic gradiometers, adaptive signal processing, and moderately shielded rooms can provide environmental noise attenuation in excess of 107. Two types of MEG signal analysis techniques will be discussed: those depending solely on prior noise cancellation (e.g., equivalent current dipole fit and minimum norm), and those intrinsically providing additional cancellation of far and near field noise (e.g., beamformers). The principles and behavior of beamformers for variations in signal and noise will be explained. Several beamformer classes will be discussed, and the presentation will conclude with examples of their clinical applications.
Signal processing in magnetoencephalography
The subject of this article is detection of brain magnetic fields, or magnetoencephalography (MEG). The brain fields are many orders of magnitude smaller than the environmental magnetic noise and their measurement represent a significant metrological challenge. The only detectors capable of resolving such small fields and at the same time handling the large dynamic range of the environmental noise are superconducting quantum interference devices (or SQUIDs). The SQUIDs are coupled to the brain magnetic fields using combinations of superconducting coils called flux transformers (primary sensors). The environmental noise is attenuated by a combination of shielding, primary sensor geometry, and synthetic methods. One of the most successful synthetic methods for noise elimination is synthetic higher-order gradiometers. How the gradiometers can be synthesized is shown and examples of their noise cancellation effectiveness are given. The MEG signals measured on the scalp surface must be interpreted and converted into information about the distribution of currents within the brain. This task is complicated by the fact that such inversion is nonunique. Additional mathematical simplifications, constraints, or assumptions must be employed to obtain useful source images. Methods for the interpretation of the MEG signals include the popular point current dipole, minimum norm methods, spatial filtering, beamformers, MUSIC, and Bayesian techniques. The use of synthetic aperture magnetometry (a class of beamformers) is illustrated in examples of interictal epileptic spiking and voluntary hand-motor activity.
Jiri Vrba, Stephen E. Robinson, Signal Processing in Magnetoencephalography, Methods, Volume 25, Issue 2, 2001, Pages 249-271, ISSN 1046-2023, http://dx.doi.org/10.1006/meth.2001.1238.
(http://www.sciencedirect.com/science/article/pii/S1046202301912381)
Movement-Related Desynchronization of the Cerebral Cortex Studied with Spatially Filtered Magnetoencephalography
Event-related desynchronization (ERD) within the α and β bands on unilateral index finger extension and hand grasping was investigated on six normal volunteers with magnetoencephalography (MEG). A novel spatial filtering technique for imaging cortical source power, synthetic aperture magnetometry (SAM), was employed for the tomographic demonstration of ERD. SAM source image results were transformed into statistical parametric images. On the same hand grasping task, a functional MRI (fMRI) study was conducted on two subjects and compared with the ERD result. When the MEG data were analyzed with the fast Fourier transformation, power attenuation within the α and β bands was evident on the contralateral sensorimotor area just prior to movement onset. The tomographic distribution of ERD was clearly obtained with SAM statistical imaging analysis. The equivalent current dipole (ECD) for the signal-averaged motor field was localized to the hemisphere contralateral to the hand movement, roughly at the center of the region displaying β-band ERD. The signal increase on fMRI roughly colocalized with the ERD on the contralateral sensorimotor area. In conclusion, with the novel spatial filtering technique for the brain magnetic field, SAM, cortical regions contributing to ERD on finger movement were successfully demonstrated in a tomographic manner. The relative colocalization of the contralateral SAM ERD with ECD as well as the fMRI activation suggests that SAM is a practically useful technique to extract event-related signals from brain noise.