Early Detection of Pediatric Seizures in the High Gamma Band

Abstract

Using data from the MIT Physionet EEG database collected at the Children’s Hospital Boston, we identify a method of detecting seizures in ten pediatric patients at least thirty seconds before seizure onset by identifying significant preictal locations and their respective frequencies within the high gamma band of 30 through 100 Hz. We analyze the potential predictive performance of event-related potential analysis in this high gamma band, provide evidence that detection algorithms should take into account the varying strength of a patient’s common frequency extremes, and provide evidence that patient-specific approaches to machine learning algorithms may be more successful in the detection of pediatric seizures as they are more difficult to detect than adult seizures. Using these results, machine learning detection algorithm performance on pediatric patient data, which is prone to issues from limitations within the algorithms, may be significantly improved by incorporating high gamma band signal processing at the locations identified by this process.

Keywords

• Biomedical engineering
• biomedical signal processing
• computational systems biology
• encephalography