Abstracts

Rationale: A prospective multicentre study is being performed at three epilepsy monitoring units (EMU) to evaluate the automatic seizure detection system EpiScan [1]. The study will determine the detection performance on a large prospective patient group and quantify the added value of an automatic seizure detection for epilepsy diagnosis and for workflow optimization. Here we will present preliminary results of the first 43 out of 180 planned patients. Methods: 43 consecutive patients over the age of 18 undergoing long term video-EEG monitoring for epilepsy diagnostic and pre-surgical evaluation were evaluated. During recording normal clinical procedures were applied. The seizure detection system was used out-of-the-box and no parameters were changed during the analysis. EEG-data were visually analysed by experienced epileptologists based on EEG and video following the routine procedures for EEG analysis. The resulting seizure markings were taken as basic truth and the results of the automatic detection were evaluated based on these seizure markers. A marker of the automatic detection was defined as true positive (TP) when it was within 3 minutes after the marker based on visual analysis. A marker outside of this time window was defined as false positive (FP). False positives lasting for more than 30 seconds were counted multiple times. A false negative (FN) is defined as missed seizure. The sensitivity is defined as the ratio of the number of TP to the number of all seizures (TP+FN). The specificity is given in false alarms per hour (FA/h) calculated as the ratio of the number of FP to the number of recorded hours. Results: Preliminarily results from the first 43 patients with a total of 3602 hours of EEG-recording are presented. Out of 43 patients 17 had seizures, 84 in total. All seizures were detected by the automatic detection system for 8 out of 18 patients. For 7 patients a sensitivity of at least 50% was achieved, for 2 patients a sensitivity of less than 50% was reached. The mean sensitivity over all patients was 73%. For the false alarm rate per hour we calculated an average of 0.32 FA/h. In addition to the detection of epileptic seizures the automated system provided several markers in the EEG that contained pathologic EEG like spike-wave complexes, slowing or theta oscillations. Conclusions: Our preliminary results show that the new seizure detection system EpiScan can be of value for a large fraction of patients reducing the time to analyse the data. The false alarm rate is low enough to be used as an online alerting system. The good performance was achieved without the need to adjust parameters on the software, which was used as shipped. In contrast to many other reports on seizure detection we compared the results of the automatic system with the result of the normal clinical workflow on consecutive patients in a prospective study. Furthermore, the automated system provided additional information by dragging attention to interictal parts of the EEG that show abnormal activity. [1] M. Hartmann et.al. "Online seizure detection for epilepsy monitoring units"; AES 2011