STFT-Based Segmentation in Model-Based Seizure Detection

To aid the review of long-term electroencephalograph (EEG), it is necessary to develop automatic seizure detection methods. In the literature, numerous seizure detection methods based on parameterization of the EEG have been presented. Recently a new patient-specific model-based method using Statistically Optimal Null Filters (SONF) has been proposed for seizure detection [1], This method uses stationary segments of a template seizure to generate the necessary seizure model (basis functions) that is used for all subsequent seizure detections. In this approach, the necessary stationary segments within the template are manually identified based on the constancy of the dominant rhythm. The manual selection of stationary segments is cumbersome in practice. In this paper, we present short-time-Fourier-transform (STFT) based automatic segmentation of template seizure resulting in practically usable model-based seizure detection. To assess the performance of the proposed algorithm, a comparison with the visual (manual) method of epoch selection on simulated as well as on the template seizures of five different patients is done. The overall performance improvements are evident in terms of enhanced seizure detection sensitivity and reduced number of false positives.