Discriminative and Generative Classification Techniques Applied to Automated Neonatal Seizure Detection

Abstract

A number of automated neonatal seizure detectors have been proposed in recent years. However, there exists a large variability in the morphology of seizure and background patterns, both across patients and over time. This has resulted in relatively poor performance from systems which have been tested over large datasets. Here, the benefits of employing a pattern recognition approach are discussed. Such a system may use numerous features paired with nonlinear classifiers. In particular, two types of nonlinear classifiers are contrasted for the task. Additionally, it is shown that the proposed architecture allows for efficient classifier combination which improves the performance of the algorithm. The resulting automated detector is shown to achieve field leading performance. A particular strength of the proposed algorithm is the performance of the algorithm when very low false detections are required, at 0.25 false detections per hour, the system is able to detect 75.4% of the seizure events.

Toward a Personalized Real-Time Diagnosis in Neonatal Seizure Detection

“…The \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}$j^{\mathrm {th}}$ \end{document} Gaussian component for class \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}$C$ \end{document} is a \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}$d$ \end{document} dimensional Gaussian function with mean vector \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}$\mu _{C,j}$ \end{document} and covariance matrix \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}$\Sigma _{C,j}$ \end{document}. The set of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}$M_{c}$ \end{document} mean vectors weights and covariance matrices then form the parameter set \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}$\theta _{C}$ \end{document} for class \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}$C$ \end{document}.In contrast to [22], the training procedure of the seizure and non-seizure models is different in this study. First, a so-called Universal Background Model (GMM-UBM) [32] is constructed by training on all the available training data: seizure, non-seizure, artefactual.…”

On sound-based interpretation of neonatal EEG

“…In order to create this database of 200 segments, a larger database which was previously used for seizure detection algorithm development was utilized [4,8]. This large database consists of long unedited multichannel EEG recordings from 18 newborns totalling 816 hours of duration with 1389 seizures From the 1389 annotated seizures, only a small fraction was annotated on a per-channel basis and rarely the whole seizure from the beginning and the end had a per-channel annotation.…”

“…However, the use of aEEG amongst the neonatal population has several limitations, waveform information is lost and its effectiveness in seizure detection varies with experience and is poor when compared to EEG [2]. Significant research has been conducted in the area of objective detection of seizure events using artificial intelligence [3][4][5][6]. These algorithms aim to provide clinicians with a support in diagnosing abnormal EEG activity, and can achieve accurate seizure detection, though no algorithm will detect all seizures or perform without any false alarms.…”

Clinical implementation of a neonatal seizure detection algorithm

“…A wider feature set which included spectral slope features from speech recognition has been examined in [29] . A Gaussian mixture model classifier has been developed in [30] and contrasted to SVM with the classifier combination performed in [31] . Adaptive spatial weighting of EEG channels based on the statistics of spatial neonatal seizure distributions has been introduced in [32] .…”