Alan S. Gevins

Surface and Deep EEG Correlates of Surgical Outcome in Temporal Lobe Epilepsy

Summary: Interictal and ictal EEG characteristics derived from limited surface montages and medial temporal lobe sites were compared with long‐term seizure relief following anterior temporal lobectomy in 52 epileptics. Patients were classified into one of four surgical outcome groups, ranging from seizure free to no clinical improvement. For each patient, interictal records were analyzed according to deep and surface spike characteristics and background activity. Ictal records were analyzed according to the proportion of episodes initiated in a unilateral or bilaterally synchronous fashion, the proportion of surface or deep onsets, the variability of onset location, and the morphology of seizure onsets. Interictal EEG variables that correlated with surgical outcome included: (a) various types of bilaterally synchronous surface/deep spikes; (b) diffuse background slowing; (c) sharp waves; and (d) the presence of multiple independent deep spike patterns in the lobe chosen for resection. Relevant ictal EEG variables included: (a) episodes initiated in a bilaterally synchronous fashion; (b) variability in seizure onset location; (c) the proportion of precisely focal onsets from deep sites; (d) the proportion of surface onsets: and (e) the proportion of onsets from the side chosen for resection. Multivariate analysis of these data with linear, stepwise, discriminate analysis and adaptive, nonlinear. distribution‐free pattern recognition demonstrated that: (a) both interictal and ictal EEG characteristics can independently predict surgical outcome at levels significantly better than chance; (b) ictal and interictal EEG data contain non‐redundant information for making such predictions; and (c) nonlinear pattern recognition techniques are capable of deriving the most accurate rules for predicting the effects of surgery.

EEG patterns during ‘cognitive’ tasks. I. Methodology and analysis of complex behaviors

This paper presents a methodology which uses nonlinear pattern recognition to study the spatial distribution of EEG patterns accompanying higher cortical functions. The multivariate decision rules reveal the essential EEG patterns which differentiate performance of two tasks. Cross-validation classification accuracy measures the generality of the findings. Using this method, EEG patterns were derived from a group of 23 adults during performance of several complex tasks, including Kohs block design, writing sentences, mental paper folding, and reading silently. These patterns discriminate between the tasks, are consistent with, and extend the results of, visual EEG interpretations and univariate analysis of spectral intensities. Since writing sentences could not be distinguished from mere scribbling, it is unclear whether the EEG patterns found to distinguish complex behaviors were related to the cognitive components of tasks, or to sensory-motor and performance-related factors.

Automated analysis of the electrical activity of the human brain (EEG): A progress report

Clinical evaluation of electroencephalographic (EEG) recordings is based on complex subjective processes of data reduction and feature extraction. The high dimensionality of the EEG signal, its variability, and the lack of standard population values have retarded development of automated systems. An interactive, real-time analysis system (ADIEEG) has been implemented to develop features to simplify visual interpretation and facilitate automated classification. It uses a 40 000 word PDP15-PDP11 dual processor computer. Resident code occupies approximately 11 000 locations, while a maximum of 12 000 locations are used for buffers. The system performs 1) continuous spectral analysis using the fast Fourier transform to produce estimates of power and coherence, 2) parallel time domain analysis to detect sharp transients significant to diagnosis, 3) several forms of graphics, 4) simple algorithms to reject noncortical and instrumental artifact, 5) interactive parameter alteration and on-line feedback to adjust decision thresholds when necessary, and 6) extraction of diagnostically helpful features using heuristics based on clinical EEG. The ADIEEG system resides in the University of California, San Francisco Medical Center, and Langley Porter Neuropsychiatric Institute.

EEG patterns during ‘cognitive’ tasks. II. Analysis of controlled tasks

This experiment was designed to distinguish possible EEG correlates of the cognitive components of tasks from EEG patterns associated with stimulus characteristics, limb and eye movements, and performance-related factors such as subjects ability and effort. Thirty-two right-handed adults each performed 30 trials, lasting 6-15 sec each, of four simplified, controlled tasks: mental rotation of geometric forms, serial addition of a column of signed digits, substitution of letters with subsequent word recognition and visual fixation. The first three tasks could not be differentiated from each other. Each of these tasks could be differentiated from visual fixation by approximately 10% generalized reductions in alpha and beta band intensities, and slight increases in theta band intensities frontally and occipitally. We conclude that the EEG patterns which differentiated the complex tasks described in Part I were due to inter-task differences in stimulus characteristics, efferent activities and/or performance-related factors, rather than to cognitive differences. With these controls, no evidence for lateralization of different types of cognitive activity was found in the EEG.

Neuropathological Findings Following Temporal Lobectomy Related to Surface and Deep EEG Patterns

Summary: Interictal and ictal characteristics of preoperative EEG recordings, derived from limited surface montages and medial temporal lobe sites, were compared with the results of pathological studies done on resected lobes obtained from 44 patients with complex partial seizures. Pathological material was divided into four groups: (a) sclerosis (mesial temporal or restricted to pes hippocampi); (b) neoplasia (mainly hamartomas); (c) miscellaneous lesions; and (d) no significant lesions. Interictal EEG correlates of no pathology included bilaterally synchronous surface spikes (with or without simultaneous deep spikes) and independent surface spikes (with or without simultaneous deep spikes) on the side of lobectomy. Ictal EEG correlates of no pathology included unilateral surface or surface/deep onsets, bilaterally synchronous surface onsets, more than one onset location, and suppression at onset. Focal onsets correlated with sclerosis. Frequent interictal spike activity in the non‐lobectomized lobe and fast buildup at onset of ictus suggested neoplasia. Many of the EEG correlates of no pathology are known to correlate with poor postsurgical seizure relief, due probably in part to the fact that absence of pathology in the resected specimen is a negative prognostic sign. Patients with sclerosis could be distinguished from patients with no demonstrable pathology with 81% cross‐validation classification accuracy using a distribution‐independent, nonlinear classifier. Both interictal and ictal EEG measures were used by the classifier, and one may conclude that ictal and interictal EEG recordings contain nonredundant information for predicting the presence and type of underlying pathology.

On-line computer rejection of EEG artifact

Abstract Simple, on-line, frequency domain procedures to detect non-continuous artifact in the waking EEG are presented. Individual algorithms detect head and body movements, large muscle potentials and eye-movement potentials. These algorithms are implemented as program modules in an interactive, real-time, spectral and transient analysis and display system, ADIEEG, described elsewhere. The systems performance in detecting artifact-contaminated EEG in 35 normal and abnormal, 3 min, 8-channel recordings was compared with that of the consensus of three expert scorers. The system correctly detected 65% of the artifact events identified by the consensus of expert scorers. Twenty-seven percent of the detections made by the system were of events that had not been marked by any of the three scorers. This performance was not statistically different from the average of the individual scorers vs. the consensus. The largest number of false detections were of intermittent, high-amplitude events of cortical origin which did not occur during the supervised calibration period.

Pattern recognition of human brain electrical potentials

Since brain electrical potentials (BEPs) are correlated with a variety of behavioral and clinical variables, especially tight experimental designs are necessary. Primary analysis, which usually consists of spectral analysis, linear prediction, or zero-cross detection, should match the time scale and dynamics of the states or processes being investigated. Nonneural contaminants must be removed from BEPs prior to computation of summary features. Principal components analysis, ad hoc methods, and stepwise discriminant analysis have been used to extract independent, intuitively appealing, and good-classifying features, respectively. Most pattern classification algorithms have been applied to BEPs including decision functions, trainable classification networks, distance functions, syntactic methods, and hybrids of the preceding. Because of its wide availability, most studies have used stepwise linear discriminant analysis.

Computer rejection of EEG artifact. II. Contamination by drowsiness

As part of an effort to automatically measure a background EEG baseline against which changes due to therapy or experimental manipulations may be measured, algorithms to detect EEG patterns associated with drowsiness have been developed and objectively evaluated. The decision of drowsiness is tentatively based upon changes in simple signal features, including increased ratios of both delta-band to alpha-band and theta-band to alpha-band spectral intensity as compared to thresholds automatically determined from a waking calibration period. Several heuristic criteria are then required to reach a final decision. Thirty-one normal and abnormal, 3-minute, 8-channel clinical EEG recordings containing drowsiness were scored by 5 expert scorers. Out of a total of 106 events labeled drowsy by at least one judge, 85 were found by a consensus of 3 or more of the 5 experts. On the 20 recordings not used for training the decision thresholds (testing data set), the system found 84% for the 85 episodes found by the consensus, and 89% of the 62 episodes found by all 5 scorers. Only one event was found by the system which was not found by any scorer, or which did not border on a consensus-defined episode of drowsiness. This performance is adequate to justify inclusion of these algorithms into a previously described real time EEG analysis system, ADI-EEG, allowing integration of the decisions of the separate subsystems for detection of artifact, sharp transients and drowsiness.

The use of brain electrical potentials (BEP) to study localization of human brain function.

Various techniques such as neuropsychological diagnosis of individuals with focal lesions, stimulation of neurosurgery patients, and regional cerebral blood flow have been used to elucidate the major anatomical and functional divisions of the human cerebral cortex. Because of insufficient spatial sampling and other limitations, only minor support for these divisions comes from brain electrical potential (BEP) experimentation. The use of EEG to localize different neuropathologies and to screen and track the evolution of seizure disorders is fairly reliable and still widely practiced. Its use, however, in localizing higher cognitive functions is much more complicated and has not stood the test of scientific scrutiny because of methodological problems. More specifically, the failure to control for the stimulus, response and performance related properties of tasks in experiments has rendered ambiguous the results of most EEG studies of higher cognitive functions. Those studies which actually controlled for these properties did not find and differences between tasks.

Heuristic real time feature extraction of the electroencephalogram (EEG)

The extremely complex nature of the electroencephalogram (EEG), and the subtle, nonquantified methods of pattern recognition used by human interpreters have made EEG analysis resistant to automation. Attempts at pattern recognition using multivariate classification procedures have not produced generalizable results due to the inadequate degree and quality of feature extraction prior to classification.n A real time, on-line EEG analysis strategy is described which incorporates feature extracting algorithms derived from models of human EEG interpretation. A system based upon this strategy has been implemented on a dedicated minicomputer. It includes: 1) spectral analysis using the Fast Fourier Transform (FFT) to produce continuous estimates of power and coherence; 2) parallel time domain analysis to detect the occurrence of sharp transient events of possible clinical significance; 3) continuous isometric display of spectral and transient functions; 4) spectral and time domain algorithms for the rejection of noncortical and instrumental artifact; 5) heuristics to isolate patterns and events of potential clinical significance; 6) interactive alteration of analysis and display parameters to facilitate manipulation of data from various experimental paradigms; 7) on-line feedback to alter, when necessary, artifact rejection, transient detection and feature extraction decision thresholds.