Human Intracranial High Frequency Oscillations (HFOs) Detected by Automatic Time-Frequency Analysis

Abstract

ObjectivesHigh frequency oscillations (HFOs) have been proposed as a new biomarker for epileptogenic tissue. The exact characteristics of clinically relevant HFOs and their detection are still to be defined.MethodsWe propose a new method for HFO detection, which we have applied to six patient iEEGs. In a first stage, events of interest (EoIs) in the iEEG were defined by thresholds of energy and duration. To recognize HFOs among the EoIs, in a second stage the iEEG was Stockwell-transformed into the time-frequency domain, and the instantaneous power spectrum was parameterized. The parameters were optimized for HFO detection in patient 1 and tested in patients 2–5. Channels were ranked by HFO rate and those with rate above half maximum constituted the HFO area. The seizure onset zone (SOZ) served as gold standard.ResultsThe detector distinguished HFOs from artifacts and other EEG activity such as interictal epileptiform spikes. Computation took few minutes. We found HFOs with relevant power at frequencies also below the 80–500 Hz band, which is conventionally associated with HFOs. The HFO area overlapped with the SOZ with good specificity > 90% for five patients and one patient was re-operated. The performance of the detector was compared to two well-known detectors.ConclusionsCompared to methods detecting energy changes in filtered signals, our second stage - analysis in the time-frequency domain - discards spurious detections caused by artifacts or sharp epileptic activity and improves the detection of HFOs. The fast computation and reasonable accuracy hold promise for the diagnostic value of the detector.

Automatic detection of high frequency oscillations during epilepsy surgery predicts seizure outcome

“…Several automatic HFO detectors have been developed by different research groups (Staba et al , 2002, Gardner et al , 2007, Worrell et al , 2008, Blanco et al , 2010, Dümpelmann et al , 2012, Birot et al , 2013, López-Cuevas et al , 2013, Burnos et al , 2014, Burnos et al , 2016b. The general implementation follows a two-stage procedure: a first step aims to identify a reliable threshold that we use to isolate events of interest (EoI), and a second step recognizing HFOs from spurious EoI, e.g.…”

A Simple Statistical Method for the Automatic Detection of Ripples in Human Intracranial EEG

“…Another option to reduce the FDR and detection of artifacts is to apply a post-processing step to eliminate falsely detected events and leave only "true" HFOs. This can be done either automatically, using an artifact rejection algorithm (Burnos et al 2014;Cho et al 2014;Amiri et al 2016;Gliske et al 2016) or data classification via clustering (Blanco et al 2010;Malinowska et al 2015), or manually with supervision by experts.…”

“…There are several means to reduce the FDR, e.g. applying post-processing steps (Burnos et al 2014;Cho et al 2014;Amiri et al 2016;Gliske et al 2016) or using human validation (Staba et al 2002;Gardner et al 2007;Crépon et al 2010). Here we propose another approach, in which α is optimized based on FDR instead of FPR.…”

“…More recently, several algorithms for automated artifact rejection have been developed (Burnos et al 2014;Cho et al 2014;Amiri et al 2016;Gliske et al 2016). These algorithms are independent from the initial detection, which is accomplished through either human visual analysis (Amiri et al 2016) or an existing detector (Burnos et al 2014;Cho et al 2014;Gliske et al 2016).…”

“…These algorithms are independent from the initial detection, which is accomplished through either human visual analysis (Amiri et al 2016) or an existing detector (Burnos et al 2014;Cho et al 2014;Gliske et al 2016). These methods reduce the number of false detections, but again increase the complexity and number of parameters, making implementation more challenging.…”