3.3. Method for Ripples and Spikes Identification

To more likely achieve some ripples and spikes, random data from interictal bipolar-montage channels placed inside epileptic zone were at first band-pass filtered from 80–500 Hz, and the root-mean-squared (RMS) value in a 3 ms moving window was computed. A sequence of the RMS values that stays above 5 SD (standard deviation) over the mean of the RMS baseline for at least 6 ms was identified as a putative HFO. Events separated by less than 10 ms were clustered together.

An HFO was confirmed to be true if the rectified band-pass filtered signal had 6 or more peaks that crossed a preset threshold (i.e., 3 SD above the mean of a rectified band-pass filtered baseline) [71].

For the study purposes, we coded the Staba 2002 algorithm (see: [71] and applied it for 2 channels randomly recorded data from 8 different patients (4–24 h); as a result, we were able to automatically detect true ripples and, unfortunately, some false positives (due to spikes), as illustrated in Figure 1, where the flowchart shows how the first step (band-pass filtering) of ripple detection algorithm causes false positive results due to spike occurrence (right side).

Flowchart—spike detection, true and false positive.

Flowchart illustrated with Figure 2 shows the steps taken for choosing the best threshold and for spike removal.

Flowchart with the steps of choosing the best threshold and removing spikes.

Finally, we used visual inspection to verify the detected ripples and spikes. Here, we considered the detected events as data set and divided into two following groups: training and testing sets. Each set has events of ripples and spikes.