53. Methods of high frequency oscillations detection: Advantages and disadvantages

Abstract

Background High frequency oscillations (HFOs) represent new electrographic marker of epileptogenic tissue and they are considered as a surrogate marker of seizure onset and epileptogenic zones. HFOs are recorded mainly in intracranial recordings. Visual analysis of HFOs in long-term recordings is extremely difficult due to the low signal-to-noise ratio of HFOs. Successful integration of HFOs into presurgical evaluation requires development of reliable methods of automatic HFO detection and quantification. We aimed to examine performance of three new HFO detecting algorithms and compared their performance with published detectors. Methods We implemented three published detectors which utilize RMS, line length or Hilbert transform approach to detect HFOs. We have developed additional three types of detectors which utilize short time energy estimation, Hilbert envelope and Bayesian evidence. All HFO detecting algorithms were applied to gold standard datasets and their performance quantified. Results Line length and Hilbert detectors detected the highest number of HFOs. The lowest number of the detections was achieved by RMS and energy estimating detectors. According to the results, the detectors can be divided into two groups. One group is characterized by high sensitivity. These algorithms detect nearly all the labeled HFOs events, but suffer from the high false positive detection rate. Second group of detectors have high positive prediction value but lower sensitivity. Our Hilbert envelope detector demonstrated the best performance of all evaluated detectors. Conclusions To improve the performance of detectors with high sensitivity will require to develop additional post-processing steps to remove the majority of false detections. Meanwhile detectors with low sensitivity will detect only high-amplitude HFOs. Future selection of the most appropriate algorithm for HFO detection in intracranial recordings will require detail understanding of the clinical significance of low-amplitude HFOs and major sources of false positive detections. Supported by Grants from IGA NT11460, NT13357, NT14489, GACR 14-02634S and Neuron Fund (NFKJ 001/2012).