Abstracts

Rationale: This study proposes an early seizure detection algorithm to be used for closed loop stimulation applications developed by the cluster of excellence BrainLinks BrainTools at the University of Freiburg. For successfully disrupting an epileptic seizure, an early detection of the seizure onset is needed. We propose the Instantaneous Weighted Power Ratio (IWPR) algorithm that was designed to detect one of the most commonly encountered seizure onset patterns in intracranial EEG: low voltage fast activity (LVFA). Methods: IWPR is a power ratio between four frequency bands. The intracranial EEG (iEEG) signal is filtered in the frequency bands of interest, and then the Hilbert transform is computed for the signal within each frequency band. IWPR is computed as the weighted sum of powers within gamma (30-126Hz) and extra slow activity (0-1Hz) divided by the sum of powers within the alpha (7-13Hz) and beta (13-30Hz) frequency bands. The IWPR feature is preprocessed to have zero mean and unitary variance and then a threshold is put at 2.75 standard deviations. An additional condition that the gradient of the low frequencies is decreasing at the moment of detection is used to remove some of the false detections. The algorithm is tested using 6 patients from the European Epilepsy Database (Freiburg University Clinic), containing a total number 1129 hours (188 hours average per patient) of iEEG recordings and a total of 60 LVFA seizures. Results: The average seizure detection rate was 96.8% (81-100% range), the average delay was 2.5 seconds (0.8 - 4.5 seconds range) and the average false detection rate was 35 per hour (26 - 40 false detection range). Conclusions: The novelty of IWPR as a feature for seizure detection and the advantage of using Hilbert transform instead of other methods is the instantaneous frequency, which is calculated for each data point. Therefore the use of time windows for computing FFT is no longer needed and no lag is inserted due to using time windows. In some cases, the low detection delays might make the difference between a successful closed loop stimulation device and a device with moderate outcomes in terms of seizure disruption. Research founded by grant number EXC 1086.