Introduction: The distinction between sharp transients and interictal epileptiform activity in routine scalp EEG is important. IFCN nomenclatures have not included quantitative morphological characteristics in the definition of epileptiform activity (Kane 2017). Reproducibility is not optimal.
Methods: From ca. 13,000 standard EEGs reported using the SCORE standard (Beniczky 2017), we extracted patient’s first EEGs with either a normal EEG with sharp transients, or an EEG supporting focal epilepsy and focal interictal epileptiform activity. We measured morphology on the first sharp SCORE-annotated graphoelement in a custom EEGLAB-based tool. Two raters measured amplitude, duration, slope, sharpness, slow wave area, frequency dissimilarity to background, and another measure of change from background. Halford’s 5-point scale on how epileptiform was also scored. We developed pseudoreference ranges. A prediction score was developed with regression, split in a development and test set. Diagnostic performance of the score was measured with ROC curves.
Results: 3123 EEGs were included, 866 of which were epileptiform. Interrater reproducibility for quantitative measures was high (𝜌>0.85). Reproducibility for Halford’s scale was low (κ=0.2). 5% and 95% pseudoreference ranges showed overlap between sharps and epileptiform activity for individual quantitative measures. After predictive model building, descending amplitude, onset slope, after slow area, age, and number of selected channels were chosen. The prediction score showed good diagnostic performance, with an area under the curve of the ROC curve of 0.83. Sensitivity at one cutpoint was 58% with a specificity of 90%.
Conclusion: Morphological characteristics overlap between sharp transients and epileptiform activity. Halford’s 5-point score had low reproducibility in our hands, while quantitative measures had high reproducibility. We provide pseudoreference ranges and a prediction score which may be helpful to colleagues in making a distinction. Our free tool could be used in research studies, and could be implemented by EEG vendors.