Abstracts

Rationale: The accuracy of interictal electric source imaging (ESI) in patients with focal epilepsy relies on the number of EEG electrodes and the number of accurately detected, averaged interictal epileptiform discharges (IED). Visual detection of IED for ESI is time-consuming, but semi-automatic detections need to correctly reproduce visual findings. Thus, we compared semi-automatic detection of IED from long-term Video-EEG-Monitoring (LTM) and short-term high-density-EEG (hdEEG) with visual detection of IED and the seizure onset zone (SOZ).

Methods: We prospectively recruited patients who participated in the PROMAESIS trial (clinicaltrials.gov/ct2/show/NCT04218812). We recruited all patients from the PROMAESIS trial who underwent both, LTM over 2 days using a standardized setup of 40 scalp EEG electrodes, and hdEEG with > 250 electrodes for usually 1 h. Patients who did not have a monofocal SOZ or their SOZ could not be delimited to one or two most relevant EEG electrodes were excluded. In LTM and hdEEG IED were marked visually and semi-automatically. For semi-automatic detection, potential IED-types were detected automatically, and these were then reviewed by expert raters. Concordance of semi-automatic detections of IED in LTM and hdEEG as well as visual detections of IED in hdEEG was compared with visually detected IED and SOZ-channels in LTM. Concordance was assessed using the 40-electrode-array of LTM with one-neighbourhood-level tolerance. Concordance was studied for the most frequent IED-type and the whole irritative zone separately.

Results: Forty-eight out of 58 patients with LTM and hdEEG met the inclusion criteria. Ten patients were excluded because of multifocal or extended SOZ. Considering the most-frequent IED-type, concordance with visual detections of IED in LTM was high for semi-automatically detected IED in LTM (90%) and visual detections of IED in hdEEG (89%) if IED were detected. In contrast, in hdEEG concordance-rates of semi-automatically detected IED with visual detections in LTM (70%) were low. The most frequently visually detected IED-type in LTM (83%) and hdEEG (78%), as well as the semi-automatically detected IED-type in LTM (89%) were typically concordant with the SOZ, whereas semi-automatically detected IED in hdEEG were less often concordant with the SOZ (68%). Considering the whole irritative zone, agreement between visual and semi-automatic detections in LTM was high (accuracy: 0.97 (±0.03), sensitivity: 0.93 (±0.16), specificity: 0.97 (±0.03), PPV: 0.7 (±0.28), NPV: 0.99 (±0.02)).

Conclusions: Semi-automatic detection of IES in LTM shows good agreement with visually detected IEDs. It thus qualifies for the detection of IED to achieve high event-count for accurate ESI. In hdEEG the most frequent IED-types from LTM can be typically detected visually, but in hdEEG semi-automatic detection performs worse, probably due to short recording durations. The most frequently detected IED-type, detected visually or semi-automatically, is typically in good agreement with the SOZ.

Funding: Please list any funding that was received in support of this abstract.: No dedicated funding.