Classification of burst and suppression in the neonatal electroencephalogram

Abstract

Fisher's linear discriminant (FLD), a feed-forward artificial neural network (ANN) and a support vector machine (SVM) were compared with respect to their ability to distinguish bursts from suppressions in electroencephalograms (EEG) displaying a burst-suppression pattern. Five features extracted from the EEG were used as inputs. The study was based on EEG signals from six full-term infants who had suffered from perinatal asphyxia, and the methods have been trained with reference data classified by an experienced electroencephalographer. The results are summarized as the area under the curve (AUC), derived from receiver operating characteristic (ROC) curves for the three methods. Based on this, the SVM performs slightly better than the others. Testing the three methods with combinations of increasing numbers of the five features shows that the SVM handles the increasing amount of information better than the other methods.

Analysis of Spectral Features of EEG signal in Brain Tumor Condition

“…All EEG records were then bandpass-filtered to 1-40 Hz as the signals below 1 Hz and above 40 Hz in EEG are generally unreliable due to low signal-to-noise ratio [22], [23]. The records were rereferenced to common average reference to approximate a reference-free recording condition [23], to minimize the artifacts [24], [25], to make channel records independent i.e., to make channel records to represent local activities [24], [26], and to provide high reliability over quantitative EEG features [27]. Finally a 10-minute, artifact-free epoch from each EEG record was retained for the analysis.…”

Design and optimisation of time-frequency analysis for multichannel neonatal EEG background features in term neonates with hypoxic ischaemic encephalopathy: characterisation, classification and neurodevelopmental outcome prediction

“…To the best of our knowledge, no study has attempted to discriminate patterns we have analysed. Reference [162] used burst-suppression patterns using spectral edge frequency (SEF95), 3 Hz power, median, variance and Shannon entropy and has shown an average area under the curve (AUC) of 94% using SVM on a database consisting of six term infants. Reference [176] used burst and normal patterns in the presence of artifact using wavelet-based statistics, -domain and nonlinear features and has shown 78% sensitivity on a database consisting of sixteen term newborns.…”

“…A large set of features is extracted from the three domains: -domain, -domain and ( , ) domain. These features are previously used for EEG signal classification [35,161,162] and describe the EEG from different perspectives. -domain and -domain features represent temporal and spectral characteristics, whereas ( , ) domain features represent non-stationary characteristics of a signal.…”

“…Studies [162,176] have extracted different prominent TFD-features to characterise neonatal EEG signals. In [29], the authors used statistical ( , ) features to classify neonatal seizure.…”

“…These classifiers are also used in neonatal EEG background patterns such as burst-suppression classification [162]. In all cases the overall average performance is generated following Table 4.6 and compared with other classifiers; see Table 4.10.…”