Statistical classification of spikes in benign rolandic epilepsy.

The occurrence of spikes during an electroencephalogram is a basic feature of Benign Rolandic Epilepsy of Childhood (BREC). In this study we addressed the question of whether the interictal spike structure is different between "typical" and "atypical" BREC patients. Atypical BREC patients are characterized by having other neurological abnormalities in addition to the typical BREC symptoms. This question is of interest given the good prognosis associated with the typical form of BREC. We analyzed data from 12 typical and 12 atypical BREC patients using ten variables to describe spike morphology and topography. The non-parametric method of "classification and regression trees" (CART) was used to detect discriminating features and construct a classification rule. In this way the location and amplitude of the spike were found to provide satisfactory discrimination, suggesting that the two groups may be affected by different epileptic processes. The study showed the effectiveness of the CART methodology in dealing with the categorical and non-normal variables that can be obtained from EEG records. It illustrates a simple but logical approach to the analysis of potentially complex topographic data.