Integrating Information from Multiple Signals for the Robust Detection of Neonatal Seizures

Barry Greene, School of Electrical, Electronic and Mechanical Engineering, University College Dublin, Ireland

Abstract
Neonatal seizures are the most common central nervous system disorder in newborn infants. Long term neurological damage and impairment may result from prolonged untreated seizures. As clinical detection of seizures in the newborn is known to be unreliable, a robust and reliable automated system would be of great clinical value. This study focused on the development of detection of neonatal seizures based on fusion of pertinent information from simultaneously acquired electroencephalogram (EEG) and electrocardiogram (ECG) data.
A dataset of 11 recordings from 9 neonates containing 633 seizure events, labelled by an expert in neonatal EEG, were recorded and analyzed. Each recording contained 7-12 channels of EEG and one channel of simultaneously acquired ECG. The seizure detection performance based on the multimodal fusion of EEG and ECG data was found superior to the performance of either the EEG or ECG unimodal seizure detection systems. On a patient-specific basis, 627 of 633 (99.05%) expert-labelled seizures were correctly detected (false detection rate: 23.64%). On a patient-independent basis, 422 of 633 (73.02%) of expert labelled seizures were correctly detected (false detection rate: 36.67%).
The multimodal combination of EEG and ECG data represents a new approach in seizure detection and a significant improvement on previous reported methods.

Not available

Back to Abstract