Evaluation of an automated spike-and-wave complex detection algorithm in the EEG from a rat model of absence epilepsy

Abstract

The aim of this prospective blinded study was to evaluate an automated algorithm for spike-andwave discharge (SWD) detection applied to EEGs from genetic absence epilepsy rats from Strasbourg (GAERS). Five GAERS underwent four sessions of 20-min EEG recording. Each EEG was manually analyzed for SWDs longer than one second by two investigators and automatically using an algorithm developed in MATLAB®. The sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were calculated for the manual (reference) versus the automatic (test) methods. The results showed that the algorithm had specificity, sensitivity, PPV and NPV >94%, comparable to published methods that are based on analyzing EEG changes in the frequency domain. This provides a good alternative as a method designed to mimic human manual marking in the time domain.

An automated, machine learning–based detection algorithm for spike‐wave discharges (SWDs) in a mouse model of absence epilepsy

“…The algorithm was trained using a dataset with 2500 putative events and labels (SWD or nonSWD) from 4 expert human scorers, using fivefold internal cross validation, and tested against the performance of 2 human scorers (S1 and S4) on multiple, unannotated (no computer information), and out-of-training data records. Although other groups have developed algorithms for the detection of SWDs, [16][17][18][19] to the best of our knowledge, no other group has designed an algorithm to mirror human confidence in scoring. Instead, it drew fuzzy, tightly interleaved boundaries similar to those of human scorers.…”

“…The algorithm also correlated well with the uncertainty of individual humans when presented with repetitions of the same waveforms, thus capturing ~77% of the ambiguity in human perception of SWD waveforms ( Figure 4C). Although other groups have developed algorithms for the detection of SWDs, [16][17][18][19] to the best of our knowledge, no other group has designed an algorithm to mirror human confidence in scoring.…”

“…Although algorithms have been developed for detection of SWDs, [16][17][18][19] there has been little development of algorithms for the detection or quantification of SWDs in mice. Although algorithms have been developed for detection of SWDs, [16][17][18][19] there has been little development of algorithms for the detection or quantification of SWDs in mice.…”

“…Thus, there is a need for the development of robust, automated methods for the detection of SWDs that allow for confidence-based scoring of events along a continuum that mirrors physiologically relevant EEG features and that matches human scoring characteristics. Although algorithms have been developed for detection of SWDs, [16][17][18][19] there has been little development of algorithms for the detection or quantification of SWDs in mice. 16 This is an important gap because rodent models are currently the main research tools for understanding the basic mechanisms of epilepsy.…”

An automated, machine learning-based detection algorithm for spike-wave discharges (SWDs) in a mouse model of absence epilepsy

“…Thus, there is a need for the development of robust, automated methods for the detection of SWDs that allow for confidence-based scoring of events along a continuum that mirrors physiologically relevant EEG features and that matches human scoring characteristics. While algorithms have been developed for detection of SWDs [16][17][18][19] , there has been little development of algorithms for the detection or quantification of SWDs in mice 16 . This is an important gap because rodent models are currently the main research tools for understanding basic mechanisms of epilepsy.…”