Geavanceerde EEG technologie in epilepsie bij kinderen

The aim of this PhD was twofold. First, we wanted to investigate whether the Plug 'n Patch system, a wearable multimodal seizure detection system, was able to detect seizures in children with therapy-resistant epilepsy. It is known that children with therapy-resistant epilepsy, and more specifically children with epileptic encephalopathy, often present with different seizure types, of which tonic seizures, atonic seizures and myoclonic seizures are the main ones. Also given that extensive investigations have already been conducted for other common seizure types, such as absences and tonic-clonic seizures, it seemed important to focus on these seizure types. In a prospective study, children with tonic, atonic or myoclonic seizure history were included. They were asked to wear the wearable detection system in parallel with 24-hour video EEG recording when they came to the epilepsy monitoring unit as part of their standard of care. Three hundred and seventy seizures, including 323 motor seizures, were measured in 49 children. Extensive analysis of all these seizures showed that especially tonic seizures were visually recognizable on the wearable system. Mainly EEG, surface EMG and ACM showed stereotypical abnormalities. As follows, all wearable signals of patients with tonic seizures were blindly annotated by a neurologist. This blind detection had a low overall performance (sensitivity 41%, PPV 9%, F1 score 14% and 0.75 false alarms per hour). However, when focusing on only nocturnal seizures lasting longer than 10 or 20 seconds, the performance increased significantly (sensitivity 66%, PPV 66%, F1 score 66% and sensitivity 62%, PPV 82%, F1 score 71% respectively). Except for three patients, no false alarms were indicated for this subset of seizures. These results were encouraging, considering that the sensitivity by parents sleeping in the same room for nocturnal seizures lasting more than 20 seconds was only 52%. This study thus allowed us to demonstrate the feasibility of using a multimodal system in a group of children with therapy-resistant epilepsy, almost all of whose patients had cognitive impairment. However, to increase the performance, and to reduce the workload for neurologists in the future, the next step was to search for a seizure detection algorithm for tonic seizures in our cohort. A support vector machine classifier could detect all seizures based on EEG, EMG and ACM after early fusion and a feature selection step, but with 24 false alarms per hour. As a result, these results are not useful in clinical practice but provide important insights for further research. The second part of this thesis focused on the complex bidirectional relationship between sleep and epilepsy. In the treatment of epilepsy, the emphasis is mainly on reducing seizures. Disturbed sleep, including both disruption of sleep architecture an sich, but also sleep disorders, are a common problem in epilepsy patients. To find out to what extent this disrupted sleep has already been studied in children with therapy-resistant epilepsy, we conducted a literature review. In this review, we looked for studies using polysomnography or questionnaires. Our study was able to show that indeed subjectively, based on questionnaires, significantly more severe sleep problems are reported in children with therapy-resistant epilepsy as opposed to children without or with well-controlled epilepsy. Also, polysomnography studies showed increased sleep fragmentation and decreased REM sleep in the therapy-resistant group. However, none of the studies we found conducted a thorough analysis of what factors actually contribute to this disturbed sleep. As children with cognitive impairment, behavioural disorders and cerebral palsy are also known to have disturbed sleep, and as many of the therapy-resistant patients have such co-morbidities, we conducted a prospective study. In it, we included both children without epilepsy, children with well-controlled epilepsy and children with therapy-resistant epilepsy. We examined subjective sleep problems using a questionnaire, sleep macroarchitecture (the classic sleep stages) and some microarchitectural factors (spindles and slow waves). Indeed, the comparison between these groups confirmed that children with therapy-resistant epilepsy, compared with those with well-controlled epilepsy, showed less REM sleep and lower sleep efficiency, as well as had fewer spindles and showed a reduced decline in the spectral power of slow waves. Furthermore, the questionnaires showed significantly more severe sleep problems in the children with therapy-resistant epilepsy compared with well-controlled epilepsy. Several epilepsy and non-epilepsy-related variables, such as EEG abnormalities, medication and co-morbidities, were investigated as possible predictors for these observations. A multivariate regression analysis showed that for the effects on REM sleep and sleep efficiency, multifocal abnormalities on EEG, taking benzodiazepines and duration of epilepsy were significant predictors. For decrease of spindles, multifocal EEG abnormalities were significant as well, as was the presence of cerebral palsy as a co-morbidity. For the effects on slow waves, benzodiazepine use, medication resistance, seizures during sleep, cognitive impairment and age were all significant. This study demonstrates that, although in clinical practice physicians are sometimes quick to point to the underlying aetiology for sleep disruption in children with therapy-resistant epilepsy and epileptic encephalopathies, epilepsy-related factors certainly contribute. Disturbed sleep in these children can thus be considered multifactorial. To further confirm the results of this study, a subsequent study followed a patient longitudinally with sequential nocturnal EEG recordings. The patient in question showed highly active nocturnal EEG activity and developed therapy resistance. The initial diagnosis of self-limiting focal epilepsy evolved into an epileptic encephalopathy of spike-wave activation during sleep and a pseudo-Lennox syndrome. At diagnosis, there was already a reduced sleep spindle density and the patient subsequently developed a lower amount of REM sleep and a reduced decrease in slow-wave spectral power over the night. When the patient finally became seizure-free and no longer showed nocturnal EEG abnormalities, all sleep parameters normalised, with even increased REM sleep time relative to diagnosis. These positive changes occurred despite aggressive treatment with three antiseizure medications, including benzodiazepines. Both studies show that interictal EEG abnormalities play a role in disrupting sleep. However, given the effect of benzodiazepines found in the cross-sectional study, the road to finding better approaches to solve this is still long. In the future, the assessment of sleep should always be included in medication trials in children with therapy-resistant epilepsy. To speed up this process, automatic analysis of sleep stages may help. To this end, we conducted a study in which we tested a neural network to score sleep in children without epilepsy, children with well-controlled epilepsy and children with therapy-resistant epilepsy. For the first two groups, the performance was similar to human sleep assessment. For the therapy-resistant group, the performance was lower. However, REM sleep scoring was reliable, so this algorithm could still be useful, after validation, to detect therapy-resistant epilepsy patients with impaired REM sleep.