Study on spike-and-wave detection in epileptic signals using t-location-scale distribution and the K-nearest neighbors classifier

TL;DR

The study tackles automated spike-and-wave detection in EEG to aid epilepsy assessment by combining a $t$-location-scale ($t$LS) statistical representation with a $k$-nearest neighbors classifier. Each EEG segment is summarized by TLS parameters $(\mu,\sigma,\nu)$ estimated via maximum likelihood, and these features feed a $k$NN decision rule to distinguish spike-and-wave from non-spike-and-wave patterns. On real data, the approach uses offline training with 192 labeled segments and online testing with 46 labeled segments across 19 channels, achieving perfect $100\%$ sensitivity and $100\%$ specificity for SWD detection. The work highlights the effectiveness of heavy-tailed TLS features for EEG pattern discrimination and points to practical potential for real-time SWD monitoring in clinical settings.

Abstract

Pattern classification in electroencephalography (EEG) signals is an important problem in biomedical engineering since it enables the detection of brain activity, particularly the early detection of epileptic seizures. In this paper, we propose a k-nearest neighbors classification for epileptic EEG signals based on a t-location-scale statistical representation to detect spike-and-waves. The proposed approach is demonstrated on a real dataset containing both spike-and-wave events and normal brain function signals, where our performance is evaluated in terms of classification accuracy, sensitivity, and specificity.

Figures (6)

• Figure 1: spike-and-wave example, we can see its symmetric and regular morphology.
• Figure 2: Spikes-and-waves examples in clinical environment. Amplitude (y-axis) in mV and time (x-axis) in sec.
• Figure 3: Scatter plots of the offline training classification in 192 dataset signals, for the t-location-scale parameters $\mu$, $\sigma$ and $\nu$ for spike-and-waves events (blue dots) and non-spike-and-waves events (red dots), showing the data dispersion of the proposed approach. In a) and c) spike-and-waves tend to have a higher scale $\sigma$, in b) non-spikes-and-waves tend to have a location $\mu$ concentrated between 0 and 100.
• Figure 4: Scatter plots in the online classification of 46 test signals, for the t-location-scale parameters $\mu$, $\sigma$ and $\nu$ for spike-and-waves events (blue dots) and non-spike-and-waves events (red dots), showing the data dispersion of the proposed approach. In a) spike-and-waves tend towards the center down, in b) the trend is not very clear, although there is a great concentration of spike-and-waves in the center down near zero, and c) spike-and-waves tend to be located towards the right and near zero.
• Figure 5: Scatter plot in offline training classification in 192 dataset signals for the t-location-scale parameters $\mu$, $\sigma$, and $\nu$, we can see the discrimination between two groups whose size is the same (96 spike-and-waves and 96 non-spikes-and-waves), label 1 for spike-and-wave and label 0 for non-spike-and-wave.