In Clinical neurophysiology : official journal of the International Federation of Clinical Neurophysiology
OBJECTIVE : Studies of high frequency oscillations (HFOs) in epilepsy have primarily tested the HFO rate as a biomarker of the seizure onset zone (SOZ), but the rate varies over time and is not robust for all individual subjects. As an alternative, we tested the performance of HFO amplitude as a potential SOZ biomarker using two automated detection algorithms.
METHOD : HFOs were detected in intracranial electroencephalogram (iEEG) from 11 patients using a machine learning algorithm and a standard amplitude-based algorithm. For each detector, SOZ and non-SOZ channels were classified using the rate and amplitude of high frequency events, and performance was compared using receiver operating characteristic curves.
RESULTS : The amplitude of detected events was significantly higher in SOZ. Across subjects, amplitude more accurately classified SOZ/non-SOZ than rate (higher values of area under the ROC curve and sensitivity, and lower false positive rates). Moreover, amplitude was more consistent across segments of data, indicated by lower coefficient of variation.
CONCLUSION : As an SOZ biomarker, HFO amplitude offers advantages over HFO rate: it exhibits higher classification accuracy, more consistency over time, and robustness to parameter changes.
SIGNIFICANCE : This biomarker has the potential to increase the generalizability of HFOs and facilitate clinical implementation as a tool for SOZ localization.
Charupanit Krit, Sen-Gupta Indranil, Lin Jack J, Lopour Beth A
Automated detection algorithm, Epilepsy, Localization, Machine learning, Ripple