Assessment of a computer program to detect epileptiform spikes

This study compares an automated spike detection program to a group of 6 electroencephaographers. Since group members varied in experience, an expertise factor was devised to weight their scoring. EEGers underscored epileptiform events on 6 records in a manner analogous to the computer's storage of EEG segments. A summation of expertise factors was determined for every event. This sum was normalized and interpreted as a probability the event would be called a spike by a given EEGer. The performance of each scorer and of the computer at different amplitude thresholds was analyzed based on this probability. Higher rated scorers identified more subtle events. Lowering the threshold of the computer program produced a comparable increase in sensitivity. The increase in total events detected by the computer was linear over the range studied. While the proportion of false positive detection increased with lowering threshold, our readers have not found a moderate number of these distracting. We conclude that the computer system, while not as specific as an EEGer, can be as sensitive and can be a reliable screening editor for large amounts of monitoring data. On balance it is more effective than an EEGer for this limited purpose.