Medical Engineering & Physics 32 (2010) 829–839 Contents lists available at ScienceDirect Medical Engineering & Physics journal homepage: www.elsevier.com/locate/medengphy Heart rate based automatic seizure detection in the newborn  O.M. Doyle a,∗ , A. Temko a , W. Marnane a , G. Lightbody a , G.B. Boylan b a Department of Electrical and Electronic Engineering, University College Cork, Ireland b Department of Paediatrics and Child Health, University College Cork, Ireland article info Article history: Received 4 January 2010 Received in revised form 19 May 2010 Accepted 23 May 2010 Keywords: Heart rate Newborn Seizure detection Patient-independent Automatic SVM abstract This work investigates the efficacy of heart rate (HR) based measures for patient-independent, auto- matic detection of seizures in newborns. Sixty-two time-domain and frequency-domain features were extracted from the neonatal heart rate signal. These features were classified using a sophisticated sup- port vector machine (SVM) scheme. The performance was evaluated on a large dataset of 208 h from 14 newborn infants. It was shown that the HR can be useful for the detection of neonatal seizures for certain patients yielding an area under the receiver operating characteristic (ROC) curve of up to 82%. On evaluating the system using multiple patients an average ROC area of 0.59 with sensitivity of 60% and specificity of 60%, were obtained. Feature selection was performed and in the majority of patients the performance was degraded. Further analysis of the feature weights found significant variability in feature ranking across all patients. Overall, the patient-independent system presented here was seen to perform well in some patients (2 out of 14) but performed poorly when tested on the entire group. © 2010 IPEM. Published by Elsevier Ltd. All rights reserved. 1. Introduction The neonatal period is one of the highest-risk periods for seizures during the human life span, which are thought to affect between 1% and 5% of newborn infants [1]. Seizures occur as a result of many underlying conditions including hypoxic-ischaemic encephalopathy, meningitis, brain haemorrhage and stroke [2]. The aetiology of neonatal seizures is the primary determinant of neurodevelopmental outcome and although there is increasing evidence that they have an adverse effect on neurodevelopmen- tal outcome [3,4]. Therefore, prompt diagnosis and treatment are vital. Newborn seizures are often very difficult to detect because of their subtle clinical expression. In many cases seizures are clinically ‘silent’, i.e. there are no clinical manifestations. Clini- cal observation of suspected seizure activity, unaccompanied by electroencephalogram (EEG) monitoring has major limitations [5]. In addition, anticonvulsant treatment can lead to electroclinical dissociation whereby the clinical manifestation of the seizures decrease even though the electrographic seizure continue [6]. Mur- ray et al. [5] compared the recognition of neonatal seizures using  This work was supported by a Science Foundation Ireland (SFI), Principal inves- tigator career advancement award (SFI/05/PICA/1836). ∗ Corresponding author. Tel.: +353 21 4903156. E-mail addresses: orlad@eleceng.ucc.ie (O.M. Doyle), andreyt@eleceng.ucc.ie (A. Temko), l.marnane@ucc.ie (W. Marnane), g.lightbody@ucc.ie (G. Lightbody), g.boylan@ucc.ie (G.B. Boylan). clinical observation by experienced neonatal staff with the ret- rospective analysis of continuous multi-channel video-EEG by a clinical neurophysiologist. A total of 526 electrographic seizure events were identified from EEG analysis, 179 (34%) of these seizure had clinical manifestations evident on the simultaneous video recording. The neonatal staff identified 177 clinically suspected seizures, however, only 48 of these events had electrographic evi- dence of seizure activity. Overall, only 9% (48/526) of electrographic seizures were accompanied by clinical signs, which were iden- tified and documented by neonatal staff and overdiagnosis was common. Given these factors and the potential for seizure-induced brain damage in newborns, the development of an accurate automatic seizure detection system is urgently needed. However, interpre- tation of the EEG requires expert reviewers who are not usually available on the required 24 h basis. Therefore, research into the development of an automated EEG surveillance system has expanded rapidly. Faul et al. [7] evaluated three well known methods by Got- man, Liu and Celka and concluded that the levels of performance achieved on continuous multi-channel EEG were insufficient for use in a clinical environment. More recently Deburchgraeve et al. [8] published an automated neonatal seizure detection system which aims to mimic the human analysis of EEG. Their approach is based on the identification of two major characteristics of seizures, repet- itiveness and change relative to the background EEG. Aarabi et al. [9] proposed a multistage knowledge-based system which selects features based on relevance and redundancy analysis and classi- fies these features using a neural network and knowledge-based 1350-4533/$ – see front matter © 2010 IPEM. Published by Elsevier Ltd. All rights reserved. doi:10.1016/j.medengphy.2010.05.010