Performance metrics for the accurate characterisation of interictal spike detection algorithms

Automated spike detection methods for the epileptic EEG are highly desired to speed up and disambiguate EEG analysis. However, it is difficult to accurately and concisely present the performance of such algorithms due to the large number of recording and algorithm variables that must be accounted for. This paper summarizes the core variables involved and presents different methods for calculating the average performance. These methods incorporate weighting factors to correct for non-ideal test cases. The factors are found to have a significant effect on the appearance of the results and the performance level that the algorithm appears to achieve. Four different weighting factors are considered and a duration divided by the number of events weighting is recommended for use in future studies.