Computer-assisted interpretation of clinical EEGs

Abstract

A multivariate pattern recognition technique has been developed, to distinguish the EEGs of patients with cerebral pathology from those of normal controls and to localize any abnormalities detected. Two methods of feature extraction have been used, power spectral density and slope descriptor analysis, together with various types of feature compression. These techniques have been evaluated on EEGs from 63 patients with proven pathology. Spectral analysis proved more reliable than slope descriptor analysis and predicted the site of cerebral pathology more accurately than did visual assessment of the EEGs. This apparent improvement over the diagnostic reliability of visual analysis in considered to justify further development and evaluation of this technique.