HFO Detector Based on EEG Mean Energy Change Detection

The main goal of this work is to support automated seizure prediction by identifying preictal states using multiple detectors of high-frequency oscillations (HFOs). Assessment of HFOs from a seizure onset zone may be facilitated using automatic HFO detectors. However, studies have shown that HFOs can be generated by physiological or pathological brain processes. Distinguishing between pathological and physiological HFOs is challenging due to a lack of discriminant information in signals’ morphology and other characteristics. Other complicating factors include a lack of consistency in definitions of HFO frequency ranges, the need for expert verification of HFO detection, and lack of robustness of automatic HFO detectors. In addition, artifacts, spatial subsampling, and filtering of spikes and their high-frequency harmonics may introduce uncertainty and make automatic HFO detection difficult. This work introduces an automatic HFO detector based on detecting changes in signal mean energy through cumulative sum computation with the aim of including it in a consensus detector. Leveraging a set of automatic HFO detectors is justified by the expectation that detection must be accurate to automatically identify changes from the interictal state to the preictal state in electroencephalogram (EEG) signals, but there is no single automatic HFO detector with a positive detection rate sufficiently close to one hundred percent and a sufficiently low false detection rate. Even though the cusum detector performed lowly it is better if combined with other methods