Optical Flow Estimation Improves Automated Seizure Detection in Neonatal EEG

Abstract

Purpose:Existing automated seizure detection algorithms report sensitivities between 43% and 77% and specificities between 56% and 90%. The algorithms suffer from false alarms when applied to neonatal EEG because of the high degree of nurse handling and rhythmic patting used to soothe neonates. Computer vision technology that quantifies movement in real time could distinguish artifactual motion and improve automated neonatal seizure detection algorithms.Methods:The authors used video EEG recordings from 43 neonates undergoing monitoring for seizures as part of the NEOLEV2 clinical trial. The Persyst neonatal automated seizure detection algorithm ran in real time during study EEG acquisitions. Computer vision algorithms were applied to extract detailed accounts of artifactual movement of the neonate or people near the neonate though dense optical flow estimation.Results:Using the methods mentioned above, 197 periods of patting activity were identified and quantified, of which 45 generated false-positive automated seizure detection events. A binary patting detection algorithm was trained with a subset of 470 event videos. This supervised detection algorithm was applied to a testing subset of 187 event videos with 8 false-positive events, which resulted in a 24% reduction in false-positive automated seizure detections and a 50% reduction in false-positive events caused by neonatal care patting, while maintaining 11 of 12 true-positive seizure detection events.Conclusions:This work presents a novel approach to improving automated seizure detection algorithms used during neonatal video EEG monitoring. This artifact detection mechanism can improve the ability of a seizure detector algorithm to distinguish between artifact and true seizure activity.

Computer vision for automated seizure detection and classification: A systematic review

“…OF provides more information about seizure semiology than FD by identifying not only the region, but also the direction of motion (Figure 2). 22 OF captures motion patterns and movement periodicity of the body during seizures, providing an information‐dense account of seizure semiology that forms an integral part of many seizure detection models 14,23,24,25,26,27,28,29,30,31,32,33,34 . Like FD, OF algorithms typically result in privacy preserving output, in this case by removing identifying facial information.…”

“…22 OF captures motion patterns and movement periodicity of the body during seizures, providing an information-dense account of seizure semiology that forms an integral part of many seizure detection models. 14,23,24,25,26,27,28,29,30,31,32,33,34 Like FD, OF algorithms typically result in privacy preserving output, in this case by removing identifying facial information. Researchers often combine OF with other motion-based or appearance-based features to improve the overall performance of seizure detection algorithms.…”

“…32,85 Combination with EEG has also been used to reduce false positive seizure detections of the EEG system in neonates. 34…”

Current and Future Uses of Continuous EEG in the NICU

“…In addition to centralized EEG interpretation, automated seizure detection is another method that might allow further expansion of cEEG use in the future. Automated seizure detection algorithms applied to neonatal cEEG have wide ranges of reported sensitivities between 43 and 81% and specificities between 56 and 90% ( 53 – 59 ). The nature of neonatal seizures as being focal and often brief make automated seizure detection in neonates particularly challenging.…”

“…With one method of automated neonatal seizure detection, the odds of detecting seizures increased with increasing seizure amplitude, duration, rhythmicity, and number of EEG channels involved ( 55 ). These algorithms are limited in their application to neonatal EEG due to artifacts and false-positive events which are frequently related to nursing care, including patting, along with respiratory artifacts, sweat artifacts, and increased rhythmicity in sleep ( 53 , 55 , 58 , 59 ). Computer vision algorithms can be applied to video recording using dense optical flow estimation to reduce false positives in automated seizure detection ( 53 ).…”