Weighted Performance Metrics for Automatic Neonatal Seizure Detection Using Multiscored EEG Data

Abstract

In neonatal intensive care units, there is a need for around the clock monitoring of electroencephalogram (EEG), especially for recognizing seizures. An automated seizure detector with an acceptable performance can partly fill this need. In order to develop a detector, an extensive dataset labeled by experts is needed. However, accurately defining neonatal seizures on EEG is a challenge, especially when seizure discharges do not meet exact definitions of repetitiveness or evolution in amplitude and frequency. When several readers score seizures independently, disagreement can be high. Commonly used metrics such as good detection rate (GDR) and false alarm rate (FAR) derived from data scored by multiple raters have their limitations. Therefore, new metrics are needed to measure the performance with respect to the different labels. In this paper, instead of defining the labels by consensus or majority voting, popular metrics including GDR, FAR, positive predictive value, sensitivity, specificity, and selectivity are modified such that they can take different scores into account. To this end, 353 hours of EEG data containing seizures from 81 neonates were visually scored by a clinical neurophysiologist, and then processed by an automated seizure detector. The scored seizures were mixed with false detections of an automated seizure detector and were relabeled by three independent EEG readers. Then, all labels were used in the proposed performance metrics and the result was compared with the majority voting technique and showed higher accuracy and robustness for the proposed metrics. Results were confirmed using a bootstrapping test.

SPINDLE: End-to-end learning from EEG/EMG to extrapolate animal sleep scoring across experimental settings, labs and species

“…Another interesting aspect would involve including more human scorers and consequently applying more principled ways of combining their knowledge (e.g. see [29]).…”

Neonatal Seizure Detection Using Deep Convolutional Neural Networks

“…Furthermore, in Table 3, the epoch-based AUC, sensitivity, and specificity, as well as event-based good detection rate (GDR) and false alarm rate per hour (FAR), are reported. 62,63 Since the output of the heuristic algorithm is not continuous, Hermite spline interpolation was used to calculate the AUC. 64 For other metrics, in order to make the comparison simpler, the threshold of the CNN-RF was chosen where the sensitivities of CNN-RF and heuristic methods are equal (the horizontal dashed lines in Fig.…”

Latent Phase Identification of High-Frequency Micro-Scale Gamma Spike Transients in the Hypoxic Ischemic EEG of Preterm Fetal Sheep Using Spectral Analysis and Fuzzy Classifiers

“…Fusion of various techniques such as autocorrelation, wavelet decomposition, and nonlinear energy operator (NLEO) has been shown to also improve the overall sensitivity of spike train seizures detection in more than 217 h of recordings from term neonates [34]. Very recent studies also show that heuristic algorithms can be developed to identify spike trains when the maximums of nonlinear energy components of the signal are compared to the background EEG activity, resulting in an overall good detection rate (GDR) of 95%, tested over 353 h recordings from 81 infants [35]. More recent work has demonstrated that a combinational scheme based on the convolutional neural networks (CNN) and random forest can help to automatically identify neonatal seizures in human babies with 77% overall accuracy [36].…”