The characterization of neonatal seizures is currently based on video/EEG recordings. EEG recordings are very useful for characterizing neonatal seizures with electrographic and clinical features that are temporally correlated (electroclinical seizures). On the other hand, video recordings are the only information source available for characterizing neonatal seizures if there is no temporal correlation between clinical and electrographic features (clinical seizures). Over the years, quantitative analysis of neonatal seizures focused mostly on EEG recordings. There have been no previous studies on automated video processing and analysis of neonatal seizures due to serious technological limitations (analog video recordings, limited computer memory, low computational speed) and the lack of engineering tools for video processing and analysis. This project will rely on recent advances in video and computer technology in order to develop automated techniques for characterizing, via video analysis, neonatal seizures of focal clonic and myoclonic type and differentiating them from other normal and abnormal behaviors. The characterization of neonatal seizures will be based on quantitative information relevant only to the seizure. This information will be extracted from video-taped seizures by combining spatiotemporal decomposition of video and temporal tracking of multiple anatomical sites. The development of computer-based seizure analysis techniques will be accomplished by training a variety of artificial neural networks to classify the quantitative information extracted from video. These techniques will be evaluated and tested on an existing library of video-taped clinical events, which include focal clonic seizures, myoclonic seizures, and normal or abnormal infant behaviors. This methodology holds potential for more refined characterization of clinical events. In addition, this study may prove that quantitative analysis of video recordings of neonatal seizures can lead to the development of a stand-alone automated seizure recognition system for use in clinical settings.
PROJECT ASPECTS
Specific Aims
The primary objective of this project is to provide novel, quantitative information regarding the behavioral characteristics of neonatal seizures. We propose to develop and thoroughly test in the clinical setting a battery of automated video processing and analysis techniques designed to facilitate the recognition of neonatal seizures. The following specific hypothesis will be tested:
Hypothesis:
Automated processing of tape-recorded video images, using a combination of spatiotemporal decomposition and two‑dimensional temporal tracking of multiple anatomical sites, can be used to classify and quantify neonatal seizures of both focal clonic and myoclonic type, and differentiate them from other normal and abnormal infant behaviors.
Background and Significance
Neonatal seizures are often the first, and, in some situations, the only clinical sign of central nervous system dysfunction. Identification of seizures in the newborn initiates a prompt evaluation for a wide range of etiologies and, whenever possible, treatment of the underlying pathological processes. In some situations, antiepileptic medication is provided to diminish the likelihood of recurrent seizures, and to lower the risk of physiologic instability during seizures. The presence of seizures may also affect prognosis, particularly with regard to neurodevelopmental sequelae and risk for certain forms of epilepsy. Thus, prompt recognition of seizures by nursery personnel is important with regard to diagnosis and management of underlying neurological problems.
Despite the importance of seizure recognition, most neonatal intensive care units and nurseries have limited resources for seizure identification. The attention of nursing personnel is distributed across a large number of newborns, many of whom are ill and require continuous bedside care. Neonatal seizures are often brief and may not be recognized since nurses and physicians cannot provide continuous surveillance of all infants at risk for clinical seizures. In addition, although neonatal intensive care unit (NICU) nurses are highly trained in many aspects of care, there is significant variability in the level of skill and experience in seizure recognition among nurses. These factors illustrate the clear need for improved seizure surveillance methods that supplement direct observation by nurses and physicians, and that are practical and economically feasible.
Early attempts to characterize neonatal seizures involved primarily bedside observation and relatively brief interictal EEG recordings. The more recent development of portable EEG/video/polygraphic monitoring techniques allows investigators to assess and characterize neonatal seizures at the bedside and permits retrospective review (Mizrahi, 1996). Recent investigations using this technique have confirmed that the majority of neonatal seizures are either electroclinical (electrographic and clinical features that are temporally linked) or clinical only (clinical features with no consistent electrographic correlate) in character (Mizrahi and Kellaway, 1987). This technique is relatively expensive, is generally used for only a few hours of monitoring, and is not routinely available in many centers. Most research involving neonatal seizures has focused on analysis of EEG features, and no investigations have used quantitative techniques to characterize clinical features. This observation contrasts with the fact that the majority of seizures in the newborn are clinically expressed, either with or without an electrographic signature. Thus, automated video processing and analysis may supplement and extend human analysis of clinical seizure behaviors, and provide new information leading to more useful classification schemes.
Video recording is typically used with synchronized EEG to analyze the characteristics of a clinical seizure after its recording (Goldensohn, 1966; Penry et al., 1975; Ives and Gloor, 1978; Binnie et al., 1981; Delgado Escueta et al., 1982; Luther et al., 1982; Mizrahi and Kellaway, 1984; Pierelli et al., 1989; Bye et al., 1990; Ives et al., 1991; Oguni et al., 1992; Rector et al., 1993). This technique is limited in terms of duration of recording and the availability of trained physicians for continuous real-time analysis. However, post‑seizure analysis in the neonate can facilitate the classification of the event as epileptic or nonepileptic, determine the type of the ictal event (e.g., clonic, tonic, myoclonic, motor automatisms, and spasms), determine the EEG localization and associated clinical features of onset and evolution (focal, generalized, multifocal, alternating, migrating, etc.), reveal the precise sequence of motor components within a single seizure, and establish the temporal relationship of the observed motor activity to EEG activity.
Most studies using video as a diagnostic tool for seizure characterization deal with problems associated with video recording itself. Synchronization has been one of the major practical obstacles for the simultaneous recording of EEG and motor activity (Ives et al., 1991). The synchronization problem has been overcome by the development of integrated recording systems capable of recording motor activity as depicted on video, along with synchronized EEG and projecting both on a split screen (Oguni et al., 1992).
A video system based upon automated analysis potentially offers a number of advantages. Infants who are at risk for seizures could be monitored continuously using relatively inexpensive and noninvasive video techniques that supplement direct observation by nursery personnel. This would represent a major advance in seizure surveillance and offers the possibility for earlier identification of neurological problems and intervention. From a research perspective, automated video processing and analysis holds great potential for refined characterization of clinical events. Traditional analysis has relied upon visual analysis and consensus among pediatric neurologists, neonatologists, and clinical neurophysiologists regarding which paroxysmal behaviors represent clinical seizures. This has contributed to controversy regarding definitions of neonatal seizures, and at times even skilled and experienced clinical neurophysiologists have held different opinions regarding whether a specific behavior represents seizure activity.
Quantitative analysis using computerized video techniques may supplement and extend human analysis, and may generate novel methods for extracting relevant information from paroxysmal behaviors. In certain types of neonatal seizure behaviors, refined analysis may shed light on specific motor activity patterns or attributes that constitute seizures, as compared to repetitive behaviors that do not represent seizures and do not have the same clinical relevance. Development of a quantitative, computerized method could lead to a more rigorous definition of neonatal seizures, and could uncover key motor signatures that are not recognized using traditional visual analysis. Specific examples include the following:
· differentiation of focal clonic seizures from other repetitive movements such as tremor or semirhythmic nonpurposeful movements,
· assessment of movement characteristics in myoclonic seizures, such as amplitude and velocity of movements, and synchrony of movements between left and right extremities.
These examples represent common occurrences in the clinical setting and automated video techniques directed at analysis of specific components of movement will contribute to a more objective and quantitative analysis. In addition, these techniques may provide the basis for further work directed towards understanding the pathophysiology of certain seizure behaviors (epileptic versus nonepileptic mechanisms), and formulating more refined capabilities regarding prediction of outcome, based upon the clinical presentation of neonatal seizures.
The Clinical Research Centers for Neonatal Seizures (CRCNS) were established by the National Institute of Neurological Disorders and Stroke (NINDS) in 1991. The overall goal for this initiative was to develop a comprehensive understanding of the clinical and EEG features, predisposing risk factors, etiology and outcome of seizures in the newborn. A comprehensive database has been created which includes detailed demographic information and maternal and infant risk factors, medical and neurological problems, neurological examinations, weekly tracking of subjects throughout hospitalization, and long term follow‑up at six, twelve, and twenty four months of age. As part of this work, bedside video/EEG/polygraphic monitoring was performed (minimum of two hours for initial study), followed by repeat one hour studies 3‑5 days after the initial seizure characterization, and at the time of discharge. Additional studies were performed whenever clinically indicated, particularly when new seizure behaviors occurred. A comprehensive database containing several hundred individual clinical seizures is available to establish a library of motor signature patterns that are characteristic of focal clonic and myoclonic seizures in the newborn. Data from bedside video/EEG/polygraphic monitoring is available on videocassettes, including video images as well as digitized signals from EEG and polygraphic recordings.
Recent developments in video processing and analysis research can facilitate the analysis of neonatal seizures. These developments have been stimulated by the transition from analog to digital television and video. This transition is expected to eliminate the boundaries between television and computer technology and to expand the use of computing devices into video processing and analysis. This integration of video processing and computer techniques represents a rich opportunity for refined characterization of clinical seizures, and provides the basis for development of automated mechanisms capable of detecting the beginning of clinical seizures.
The linkage of sophisticated video and computer processing with seizure recognition and characterization represents an innovative approach which has never been used before. Development of these techniques can lead to refined characterization of repetitive motor behaviors, and differentiation of certain clinical seizures from other abnormal paroxysmal behaviors not due to seizures. Secondly, this approach will stimulate the development of important clinical applications such as automated seizure detection systems. The seizure recognition system described in this proposal will be specifically designed to identify focal clonic and myoclonic seizures. The proposed approach may not be suitable for other types of seizures involving subtle movements of body parts other than the extremities, such as ocular and orobuccolingual seizures. Nevertheless, focal clonic and myoclonic events constitute a large proportion of seizures observed in neonates. With further development and refinement of these techniques, systems which identify more subtle seizure types may be feasible