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

OBJECTIVE

Manual detection of spike-wave discharges (SWDs) from electroencephalography (EEG) records is time intensive, costly, and subject to inconsistencies/biases. In addition, manual scoring often omits information on SWD confidence/intensity, which may be important for the investigation of mechanistic-based research questions. Our objective is to develop an automated method for the detection of SWDs in a mouse model of absence epilepsy that is focused on the characteristics of human scoring of preselected events to establish a confidence-based, continuous-valued scoring.

METHODS

We develop a support vector machine (SVM)-based algorithm for the automated detection of SWDs in the γ2R43Q mouse model of absence epilepsy. The algorithm first identifies putative SWD events using frequency- and amplitude-based peak detection. Four humans scored a set of 2500 putative events identified by the algorithm. Then, using predictors calculated from the wavelet transform of each event and the labels from human scoring, we trained an SVM to classify (SWD/nonSWD) and assign confidence scores to each event identified from 60, 24-hour EEG records. We provide a detailed assessment of intra- and interrater scoring that demonstrates advantages of automated scoring.

RESULTS

The algorithm scored SWDs along a continuum that is highly correlated with human confidence and that allows us to more effectively characterize ambiguous events. We demonstrate that events along our scoring continuum are temporally and proportionately correlated with abrupt changes in spectral power bands relevant to normal behavioral states including sleep.

SIGNIFICANCE

Although there are automated and semi-automated methods for the detection of SWDs in humans and rats, we contribute to the need for continued development of SWD detection in mice. Our results demonstrate the value of viewing detection of SWDs as a continuous classification problem to better understand "ground truth" in SWD detection (ie, the most reliable features agreed upon by humans that also correlate with objective physiologic measures).