Enhancement of deep epileptiform activity in the electroencephalogram by adaptive spatial filtering (1997)
Type of ContentTheses / Dissertations
Thesis DisciplineElectrical Engineering
Degree NameDoctor of Philosophy
PublisherUniversity of Canterbury. Electrical and Electronic Engineering
AuthorsWard, Donna-Mareeshow all
The detection of epileptiform discharges (EDs) in the electroencephalogram (EEG) is an important component in the diagnosis of epilepsy. However, when the epileptogenic source is located deep in the brain, the EDs at the scalp are often masked by more superficial, higher amplitude EEG activity. Currently, these can only be investigated through invasive electrophysiological investigations such as sphenoidal electrodes, depth electrodes, and electrocorticography, or extremely expensive procedures such as magnetoencephalography. This thesis reports on a comprehensive study of the enhancement of signals from deep sources in the brain suspected of being EDs by way of a non-invasive technique - adaptive 3-d spatial filtering or, more commonly, beamforming. Initially, the EEG and the various recording methodologies are reviewed, including the neurophysiology of the brain, the mechanisms responsible for the generation of the action potential, and the anatomical structure of the cerebral cortex. The characteristics and morphology of both normal background EEG activity and abnormal EEG activity are introduced. Different recording procedures as well as signal processing of the EEG plus other procedures used in the detection and localization of epileptiform activity are described. This is followed by a review of adaptive signal processing techniques and adaptive algorithms. The beamformer, which incorporates an adaptive filter, is described along with its application to EEG signals. A forward 3-dimensional 3-layer homogeneous spherical model was applied to predict the potential distribution recorded on the surface of the scalp from a dipole source. This was used to set the beamformer's spatial response constraints. The beamformer adapts, using the least-mean square adaptive algorithm, to reduce signals from sources distant from a particular location in the brain. The location may be based on clinical evidence of a brain lesion or other prior knowledge. The beamformer produces three outputs, the orthogonal components of the source estimated at the assumed location. Simulations were performed by using the same forward model to superimpose realistic EDs on normal EEG recordings. An enhancement ratio (ER), the improvement in signal-to-noise ratio (SNR) in the beamformer output from that observed at any of the scalp electrodes, was used as a performance measure. The simulations showed the beamformer's ability to enhance signals emanating from epileptogenic foci by a mean 120% (-30% - 360%). They also indicated that the beamformer's performance is moderately dependent on the number of electrodes and EEG background but largely independent of the amplitude of the ED. The degree of enhancement is moderately dependent on eccentricity for radially and tangentially oriented dipoles. The beamformer was relatively independent of variations in polar angle location for radial dipoles. Likewise, the beamformer was relatively independent of variations in azimuthal angle location for both radially and tangentially oriented dipoles. The beamformer performs better when the dipolar source is not located directly under electrodes on the scalp. Since the beamformer was relatively insensitive to depth and to inaccuracies in estimate of location (depth and angular location) of both radially and tangentially oriented dipoles then this indicates that the beamformer is likely to be most suitable as a non-invasive means of enhancing activity from epileptogenic foci whose location is ill-defined. A clinical evaluation of the performance of the beamformer on real EEG records was also undertaken. This comprised EEG records from 11 patients, 4 of which were graded globally by an EEGer as containing epileptiform activity and 7 as containing questionable epileptiform activity. From the first group, 12 events graded as definite or questionable EDs and from the second group, 27 events graded as questionable EDs, were selected to assess the beamformer's ability to enhance real EDs. A thresholded mean enhancement ratio (ERt), the average improvement in SNR in the projected beamformer output calculated over all channels for which both input and output are greater than 0.7, was the performance measure used in all clinical data cases. The beamformer was found to enhance the 7 definite EDs by a mean 102% (range 54% - 215%) with no significant difference between referential data and bipolar data. Questionable EDs were enhanced by a mean 78% (range 3% - 259%). The referential and bipolar input and beamformed projected output records of all definite and questionable EDs were randomized and shown to an EEGer for grading. As a result of beamforming, several events were upgraded from questionable to probable, indicating that they were likely to be genuine EDs, while a few were downgraded from probable to non-epileptiform indicating that they were likely to be artifacts. Unfortunately, for the current study, there was no independent means (such as depth electrodes recordings or an EEGer panel) by which the validity of this upgrading - and hence the clinical effectiveness of the beamforming technique - could be determined. Finally, simulations investigating the performance of different adaptive algorithms in the beamformer were performed. Improvement in the performance of the LMS algorithm by a mean 31% (p>0.10) was gained by the use of a learning rate optimized to each patient's EEG. A further mean increase of 14% (p>0.20) was provided by the RLS algorithm. The simulations indicated that least squares algorithms could provide slightly better enhancement than gradient-search algorithms at the cost of being more computationally expensive. In summary, the simulation and clinical results demonstrated that the beamforming technique has considerable potential for application in clinical neurophysiology. The beamforming technique can non-invasively enhance epileptiform activity from deep foci in the brain, such as occur in mesial temporal lobe epilepsy, thereby reducing the need for alternative invasive electrophysiological investigations.