Abstracts

Rationale: Defining seizure onset is essential in chronic intracranial EEG monitoring (ECoG) for patients with refractory focal epilepsy. Visual judgement by experienced ECoG readers is usually the gold standard. We developed a frequency analysis algorithm to determine seizure onset in a more objective way and tested its performance. Methods: Spontaneous clinical seizures from 21 patients were analyzed. Seizure onset was defined as a sustained rhythmical change in the ECoG followed by a typical clinical seizure. The frequency analysis algorithm detects abrupt shifts from low to high frequency band power. The selected high frequency bands were high beta (21-28 Hz) and gamma (28-120 Hz). The algorithm included a threshold to select significant electrodes. Three experienced ECoG readers, independently and by consensus, a posteriori agreed on the seizure onset for each patient’s seizures. Multiple electrodes involved in the first 2 seconds of ECoG seizure onset were marked by the algorithm and by the three human observers, and labelled according to predefined anatomical regions. Thus seizure onset regions were defined. A comparison was made between the regions indicated by the human observers and by the algorithm. Results: Of the 21 patients the same onset region was indicated in 14 (67%). In 5 (24%) the algorithm found one (10%), two (10%) or three (5%) extra regions. In 2 (10%) the human observers found one and two regions more. In 24% of patients, the algorithm indicated exactly the same electrodes as the human observers. This was the case when only a few (1-4) electrodes were involved. Conclusions: The frequency analysis algorithm can be used to determine seizure onset and may act as an independent observer. The algorithm assists in quantifying the complexity of seizure onset in the ECoG.