Automatic detection of high-frequency oscillations in invasive recordings

High-frequency oscillations (HFOs) represent relatively new electrographic marker of epileptogenic tissue. It is starting to be used in presurgical examination to better plan surgical resection and to improve outcome of epilepsy surgery. Development of new techniques of unsupervised HFOs detection is required to further investigate the role of HFO in the pathophysiology of epilepsy and to increase the yield of presurgical examination. In this study we applied an envelope distribution modelling technique on experimental and human invasive data to detect HFOs. Application to experimental microelectrode recordings demonstrated satisfactory results with sensitivity 89.9% and false positive rate 2.1 per minute. Application of this algorithm to human invasive recordings achieved sensitivity 80%. High numbers of false positive detections required utilization of postprocessing steps to eliminate the majority of them. This study shows that envelope distribution modelling represents a promising approach to detect HFOs in intracranial recordings. Advantages of this approach are quick adjustments to changes in background activity and resistance to signal nonstationarities. However, successful application to clinical practice requires development of secondary processing steps that will decrease the rate of false positive detections.