Original ArticleAutomatic Detection of Childhood Absence Epilepsy Seizures: Toward a Monitoring Device
Introduction
Childhood absence epilepsy is a common idiopathic generalized epilepsy syndrome [1]. It is manifested in the electroencephalogram as paroxysms of high-amplitude, bilateral synchronous, symmetric, approximately 3 Hz spike-wave patterns on an otherwise normal background. A very close correlation generally exists between paroxysms of more than 2 seconds in duration and the clinical appearance of absences characterized by interruptions of intentional behavior, impaired consciousness, and for some, a blank stare accompanied by lip smacking, upward gaze, or eye blinking. Childhood absence epilepsy presents in children between ages 4 and 10 years, peaking at ages 6-7 years. A strong genetic predisposition is evident, with occurrence more often in girls than in boys. The very frequent absences (several to hundreds a day) exert a negative impact on an otherwise normal child. Untreated children often exhibit learning and attention difficulties because of their alterations of consciousness [2].
Children diagnosed with childhood absence epilepsy usually demonstrate favorable long-term outcomes. Approximately 90% of patients become seizure-free on antiepileptic drugs [3], [4], and for 65%, a total remission after 1-2 years of seizure freedom is achieved [5].
The pediatric neurologist's main objective is to neutralize all absences as long as the side effects do not exert too detrimental an effect. The goal of zero absences is motivated by cognitive issues and physical safety [3], [6]. Moreover, for those reasons, scientists have studied the advantages of ambulatory electroencephalogram monitoring for better drug dosages [7], [8], [9]. Ambulatory electroencephalogram monitoring has not proved unequivocally better than routine electroencephalograms at establishing control of absences. However, ambulatory electroencephalogram monitoring does offer a broader, better understanding of the frequency and duration of paroxystic electroencephalogram activity, compared with results in the clinical setting or according to parental report. Comparisons of a parent’s history of a patient’s seizures, observations by nurses, intensive observations by a trained observer, routine electroencephalogram results with hyperventilation and photic stimulation, and 12-hour telemetered electroencephalography indicated that 12-hour telemetered electroencephalography was the most reliable [10]. Reportedly, parents acknowledge only 6% of daytime paroxysms lasting longer than 3 seconds [11].
In the protocol for diagnosing childhood absence epilepsy and deciding on relevant drugs and doses, a patient’s history is considered together with an approximately 30-minute standard scalp electroencephalogram. To provoke an absence seizure, the child is usually asked to hyperventilate or is exposed to intermittent photic stimulation. This examination involves two shortcomings: it does not cover circadian variation, and it fails to investigate how profuse the seizures are in a normal daytime setting. On the other hand, a clinical evaluation provides the means to obtain high-quality electroencephalogram data, with few artifacts and the possibility of performing a video electroencephalogram. To cover circadian variations, patients could be admitted to an epilepsy monitoring unit, but this approach is considered too expensive and cumbersome in light of the limited benefits.
An ambulatory long-term electroencephalogram involves the disadvantage of creating a huge amount of data, requiring extensive resources for examination. In the 1970s and 1980s, multiple research groups sought to create an algorithm that would automatically recognize paroxystic activity in electroencephalogram recordings of children with absence seizures [12], [13], [14], [15]. With the methods available, they did not reach sensitivities higher than 80% and false detection rates below 2/hour. These results were deemed too faulty for use in a clinical setting.
However, the automatic detection of seizures has remained a popular subject in the literature, and many authors appear unwilling to acknowledge the very different electroencephalographic morphologies dependent on type of epilepsy and seizures. Often a certain method of detecting seizures is reported to perform with a given sensitivity and specificity for epilepsy in general, although the performance is most likely accurate only for the seizures under investigation.
We investigate how well an algorithm for automatic seizure detection can perform if we limit the target group to patients with childhood absence epilepsy. The paroxystic electroencephalogram patterns of these patients are very distinct, and little intrapatient and interpatient variability is evident [16].
If patients, and especially children, are to wear an ambulatory electroencephalogram monitoring device, it should be comfortable, discreet, robust, and safe to use. The optimal solution would involve a device ready to use in any patient without need to tweak different parameters or train the algorithm in seizure patterns before putting it into operation. To meet these requirements, we chose to focus on the design of a biomedical signal-processing algorithm that is generic (i.e., works on all patients without patient-specific optimization) and able to detect paroxystic activity from only a single electroencephalogram channel. The generic nature of the device makes it much easier for the physician to apply, and the use of only a single electroencephalogram channel reduces the need for numerous annoying electrodes, wires, and demands for large recording devices.
Section snippets
Clinical data
Standard electroencephalogram recordings from 20 patients (13 female), diagnosed with childhood absence epilepsy, were used for training and testing an algorithm for the automatic detection of seizures. The children’s mean age was 7.5 years, with a standard deviation of 1.8 years. Nineteen electroencephalogram channels were acquired according to the international 10/20 system with Cadwell Easy II (Cadwell Laboratories, Inc., Kennewick, WA) (18 patients) or Stellate Harmonie (Stellate Systems,
Results
A typical paroxysm recorded from channel F7-Fp1 is presented in Fig 1, together with its wavelet decompositions. Comparing the decompositions with the original signal, decomposition d1-d3, containing the high frequencies, describes primarily the spikes, whereas decomposition d5-d6 describes the wave portion of the paroxysm. This result indicates that investigating all decomposition levels could be valuable, despite previous reports on the insignificance of detail levels containing high
Previous findings
Several recent studies focused on the detection of absence seizures [20], [26], [27], [28]. Adeli et al. [20] analyzed the frequency content and time-related course during two absence seizures. They concluded that the wavelet transform based on a Daubechies 4 mother wavelet is appropriate to describe the spike-wave electroencephalogram signal, and only frequencies below 30 Hz are clinically relevant. Subasi [26] analyzed four channels of electroencephalograms from five patients with absence
References (38)
- et al.
The utility of ambulatory EEG monitoring in typical absence seizures
Brain Dev
(1991) - et al.
Ambulatory cassette EEG in absence epilepsy
Pediatr Neurol
(1987) - et al.
High-speed automatic analysis of EEG spike and wave activity using an analogue detection and microcomputer plotting system
Electroencephalogr Clin Neurophysiol
(1980) - et al.
Automatic detection of spike-and-wave bursts in ambulatory EEG recordings
Electroencephalogr Clin Neurophysiol
(1985) - et al.
Clinical evaluation of a method of quantification of generalized spike-wave EEG patterns by computer during prolonged recordings
Comput Biomed Res
(1977) - et al.
Patient-specific seizure onset detection
Epilepsy Behav
(2004) - et al.
Analysis of EEG records in an epileptic patient using wavelet transform
J Neurosci Methods
(2003) Automatic detection of epileptic seizure using dynamic fuzzy neural networks
Expert Syst Appl
(2006)Application of adaptive neuro-fuzzy inference system for epileptic seizure detection using wavelet feature extraction
Comput Biol Med
(2007)Algorithm architectures for patient dependent seizure detection
Clin Neurophysiol
(2006)
Drug withdrawal in patients with epilepsy: Prognostic value of the EEG
Seizure
Childhood absence epilepsy and electroencephalographic focal abnormalities with or without clinical manifestations
Seizure
EEG in childhood absence epilepsy
Seizure
Interictal paroxysmal EEG abnormalities in childhood absence epilepsy
Seizure
Proposal for revised classification of epilepsies and epileptic syndromes
Epilepsia
The neuropsychology of pediatric epilepsy and antiepileptic drugs
Epidemiology of absence epilepsy. III. Clinical aspects
Acta Paediatr Scand
Absence epilepsies
Epilepsia
Long-term prognosis of typical childhood absence epilepsy: Remission or progression to juvenile myoclonic epilepsy
Neurology
Cited by (28)
Automated seizure detection systems and their effectiveness for each type of seizure
2016, SeizureCitation Excerpt :The data from scalp EEG had a sensitivity of 74–99% and an FDR of 0.28–1/h, with an anticipation time of 16 min [28,32,36,39,40]. A device for detection of absence seizures achieved a sensitivity of 95%, with an FDR of 0.037/h, when tested in an ambulatory setting in one patient [41]. The electrodes were placed in F7-Fp1, as this yielded the best results in a previous study [37].
Non-EEG based ambulatory seizure detection designed for home use: What is available and how will it influence epilepsy care?
2016, Epilepsy and BehaviorCitation Excerpt :None of the studies evaluated EEG-based automatic seizure detection devices intended for in-home use. All studies used extracerebral signals including acceloremetry (n = 7), mattress-based sensor (n = 4), heart rate (n = 2), sound (n = 1) or video (n = 1), a combination of accelerometry and EDA (n = 1), and EMG (n = 1) (Table 1) [12–17,19–25,32,33]. All studies were performed in a hospital setting.
Subdural to subgaleal EEG signal transmission: The role of distance, leakage and insulating affectors
2013, Clinical NeurophysiologyCitation Excerpt :The spatial sampling of 1 cm interelectrode distance may leave out important generators. Bullock et al., 1995; Duun-Henriksen et al., 2012, have previously shown that the coherence and correlation between intracranial channels are high for very short distances (<10 mm). Therefore we believe that cortical generators between electrodes radiate most of their electrical discharges to the nearest intracranial electrodes.
A Logic-Based Framework Leveraging Neural Networks for Studying the Evolution of Neurological Disorders
2019, Theory and Practice of Logic Programming