Automatic detection of spike-and-wave bursts in ambulatory EEG recordings

The spike-and-wave detection scheme described in this report is based on the recognition of groups of spikes and sharp waves with loosely defined temporal and inter-channel relationships; presence of a slow wave is required only with spikes of low amplitude. Particular attention is paid to artefacts. This method allows the detection of classical 3/sec spike-and-wave activity as well as irregular patterns. The analysis produces a standard EEG tracing containing only the detected bursts, allowing conventional visual examination by an electroencephalographer. After interactive editing of false detections, a quantitative display of burst distribution can be obtained. The rate of false detections due to artefacts or non-epileptiform patterns was evaluated on eight 20 h cassette recordings; an average of 2.4-10 false detections were made per hour. When tested against hand-scoring by two EEGers of eight 2 h recordings which included very irregular spike-and-wave patterns, the computer detected 70% of the spike-and-wave activity identified by both readers (i.e., 30% false negatives), while 12% of the computer detections were not identified by either reader (false positives). The context in which such a computer method with imperfect performance can be clinically useful is discussed.