Figure 6.

External validation: application of templates to classify TLE subjects vs controls. (a) Example of a template for which the 99.95th percentile values of likelihood scores discriminate between patients with TLE and control subjects also in the case when the analysis was restricted to those with normal and normal-variant scalp EEGs (grey background). (b) Range of AUC curves for all studied templates and likelihood score 99.95th percentile-based discrimination shown together with a ROC curve derived from clinical review of scalp EEGs (AUC = 0.58). At high-sensitivities template-based classification outperforms visual scalp EEG analysis. Both approaches perform comparably at high specificities. (c) Dependence of AUC values on the fixed percentile threshold. Performance of the studied set of templates was stable at AUC ~ 0.7 and significantly higher than AUC = 0.5 throughout most of the range.