Validation of an automated seizure detection algorithm for term neonates

Sean R Mathieson,Nathan J Stevenson,Evonne Low,William P Marnane,Janet M Rennie,Andrey Temko,Gordon Lightbody,Geraldine B Boylan

doi:10.1016/j.clinph.2015.04.075

Abstract

ObjectiveThe objective of this study was to validate the performance of a seizure detection algorithm (SDA) developed by our group, on previously unseen, prolonged, unedited EEG recordings from 70 babies from 2 centres. MethodsEEGs of 70 babies (35 seizure, 35 non-seizure) were annotated for seizures by experts as the gold standard. The SDA was tested on the EEGs at a range of sensitivity settings. Annotations from the expert and SDA were compared using event and epoch based metrics. The effect of seizure duration on SDA performance was also analysed. ResultsBetween sensitivity settings of 0.5 and 0.3, the algorithm achieved seizure detection rates of 52.6–75.0%, with false detection (FD) rates of 0.04–0.36FD/h for event based analysis, which was deemed to be acceptable in a clinical environment. Time based comparison of expert and SDA annotations using Cohen’s Kappa Index revealed a best performing SDA threshold of 0.4 (Kappa 0.630). The SDA showed improved detection performance with longer seizures. ConclusionThe SDA achieved promising performance and warrants further testing in a live clinical evaluation. SignificanceThe SDA has the potential to improve seizure detection and provide a robust tool for comparing treatment regimens.

Highlights

The concept of ‘‘neuroprotective’’ intensive care has reached neonatal units worldwide, in part driven by the results of randomized controlled trials showing that therapeutic hypothermia is beneficial for term babies with a recent hypoxic-ischaemic injury (Glass et al, 2011)
In this study, comparing EEG evidence of seizures to the seizure detection acumen of NICU staff based on clinical evidence alone, of 526 EEG seizures, only 179 (34%) had any clinical accompaniment, overdiagnosis was common with only 48 of 177 (27%) clinically suspected events accompanied by EEG seizures such that only 48/526 (9%) of EEG seizures were correctly identified by clinical observation
We have previously reported the performance of our neonatal seizure detection algorithm (SDA) on a set of 17 seizure babies recorded at Cork University Maternity Hospital (CUMH), Ireland (Temko et al, 2011a) using a ‘leave one out’ (LOO) cross validation method of analysis, whereby the data of one patient is used for testing and the others used for training the algorithm and the process is repeated for each patient and the mean result reported

Summary

Introduction

The concept of ‘‘neuroprotective’’ intensive care has reached neonatal units worldwide, in part driven by the results of randomized controlled trials showing that therapeutic hypothermia is beneficial for term babies with a recent hypoxic-ischaemic injury (Glass et al, 2011). The inaccuracy of clinical recognition of seizures was demonstrated by Murray et al (2008). In this study, comparing EEG evidence of seizures to the seizure detection acumen of NICU staff based on clinical evidence alone, of 526 EEG seizures, only 179 (34%) had any clinical accompaniment, overdiagnosis was common with only 48 of 177 (27%) clinically suspected events accompanied by EEG seizures such that only 48/526 (9%) of EEG seizures were correctly identified by clinical observation. Amplitude-integrated EEG (aEEG) is widely used in NICUs for seizure detection but has been shown to perform poorly (Rennie et al, 2004). In this study seizure detection by four non-experts using CFM traces at 3 paper speeds were compared against simultaneous EEG in 19 babies.

Methods

Results

Conclusion