A comparative study between three epileptic high frequency oscillations detection strategies

In the past few years, visual detection of High Frequency Oscillations (HFOs) has led to a good understanding of the fundamental neural mechanisms underlying epileptic phenomena. However, visual marking of HFOs is very tedious, extremely time consuming and remains a subjective process. In addition, some factors like low signal-to-noise ratio and presence of artifacts in electroencephalographic signals make scoring of HFOs a difficult task. Therefore, automatic HFOs detectors have been developed by different research groups with the goal of providing reliable detection results. However, no common dataset was used to evaluate the performance of these detectors. The aim of the present study is thus to provide a fair comparative study of three existing HFOs detectors. These detectors are respectively based on the root mean square, the short time line length and the Hilbert transform techniques. This fair comparative study is performed using a common dataset with the same expert visual marking. The behavior of the considered HFOs detectors is evaluated in terms of the sensitivity, the precision, the false discovery rate and the area under the ROC curve. Our results show that the Hilbert transform based method is the most efficient and reliable method for detecting HFOs compared to the other ones.