Abstracts

Rationale: Patients with Focal cortical dysplasia (FCD) often suffer from intractable epilepsy. High frequency oscillations (HFOs) are linked to epileptic areas and to the seizure onset zone (SOZ). The removal of HFO-generating brain tissue correlates with a good postsurgical outcome. Two types of HFO patterns were observed in the ripple range: pattern 1 was defined as channels with a continuously oscillating baseline with intermittently occurring higher voltage HFOs, pattern 2 channels had a flat baseline with low frequencies and intermittent very high voltage HFOs (Figure I). This study analyzes the correlation between different patterns of HFOs and the SOZ. Moreover it investigates whether the pattern of HFOs influences the quality of automatic detection of HFOs by comparing detections based on a radial basis function (RBF) neural network with visual markings of both patterns.Methods: Consecutive patients with FCD that were recorded with intracranial grid electrodes with a 450Hz low pass filter and a sampling rate of 1024Hz were included. Ripples (80-200Hz) and fast ripples (200-450Hz) were visually identified by two independent reviewers during 3 minutes of slow wave sleep. Channels were visually classified into two different patterns of HFOs which were compared to channels inside and outside the SOZ. A RBF neural network was used to detect HFOs automatically in the same EEG-segment. The concordance between rates of visually and automatically detected HFOs was measured using Cohen s Kappa. A one-way ANOVA was used to compare Kappa values between channels with pattern 1 vs. those with pattern 2.Results: Twenty-two patients were included. Visually identified as well as automatically detected rates of HFO were significantly higher inside than outside the SOZ. Inside the SOZ, the percentage of channels that were classified as pattern 2 was significantly higher than of pattern 1 channels (62.6% 29.4% vs. 37.4% 29.4%,p=0.04). The agreement between visual and automatic detection measured by Cohen s Kappa was significantly higher in channels with pattern 2 than with pattern 1 (0.38 0.23 vs. 0.32 0.16, p<0.01).Conclusions: HFOs are significantly correlated with the SOZ in patients with FCD. HFO patterns occurring in channels with non-oscillating baselines (pattern 2) were more closely linked to the SOZ than those occurring in continuously oscillating baselines (pattern 1). Our results suggest that pattern 1 predominantly represents physiological activity whereas pattern 2 reflects epileptogenicity. The RBF neural network has greater difficulty in detecting HFO in channels with pattern 1 most likely due to the smaller difference in amplitude between HFOs and baseline than in channels with pattern 2. Regarding the distribution of HFO patterns in the SOZ we assume that pattern 1 has a smaller impact for the identification of the SOZ and deficits in detecting HFOs in this pattern are of lesser clinical relevance. Further studies are needed to investigate whether the occurrence of HFO patterns is restricted to FCD or can be seen in other causes of epilepsy.