Automated rule based graded analysis of ambulatory cassette EEGs

We describe algorithms, developed on a PDP-11/73 microcomputer, which identify spikes/sharp waves (STs), spike-and-wave complexes (SSWs), artifacts and background activity in 4-channel ambulatory EEGs. The algorithms were trained using 40 database segments. Time domain/mimetic methods were used and semantic rules, based on morphology and multi-channel contextual information, were developed to mimic the principles used in visual interpretation. The likelihood of STs/SSWs being genuine was graded from 10 to 1. This approach avoids forced classification of each event as genuine ST/SSW or not. The algorithms were then evaluated using 60 independent segments. STs/SSWs graded greater than 7 had significantly higher probability (P less than 0.005) of being genuine than those graded less than or equal to 7. Less than 4% of STs/SSWs identified by both electroencephalographers were missed. None was distinct. All 113 artifacts resembling STs/WWs were graded less than or equal to 7. Classification of 969/1117 (86%) waves in the background matched that of one electroencephalographer. The algorithms can be extended to 8- or more-channel EEGs.