Comparison of traditional reading of the EEG and automatic recognition of interictal epileptic activity

Abstract

This paper describes the evaluation of a method which has been developed for the automatic recognition and quantification of interictal epileptic activity in the human scalp EEG, using a small laboratory computer. Since it is very difficult to obtain non-EEG information clearly localizing an epileptic focus, the evaluation was performed by determining the unequivocal errors and by comparing in the interpretation of the computer display to the traditional interpretation of the paper record. Sections of 2 min of EEG which should not contain any epileptic activity (from 30 normal subjects and 30 non-epileptic patients) were analyzed, yielding the very low detection rate of 0.2 spikes or sharp waves/min/16 channels. A comparison was then performed, using data from 50 epileptic patients, between the tradiational interpretation and the interpretation of the computer display; the latter included the number of spikes and sharp waves found in every channel and the phase relationships between channels. Each of two EEGers interpreted independently the paper record and the computer display and encoded their findings regarding the localization of the focus and the extent of the abnormality in a structured report. A quantitative comparison of structured reports indicated correlation of 58% for EEGer A and 61% for EEGer B between paper record and computer display, 72% between EEGers A and B for the paper records and 84% between EEGers A and B for the computer displays. This shows that the automatic and traditional interpretations give comparable results but that the automatic method allows for more consistent interpretations between electroencephalographers.