We are currently working on a high-performance automated seizure detector. The research phase for seizure detection is well advanced and we already obtained promising results.
The seizure detection accuracy of the prototype was determined on a reference database containing 42 EEG signals from non-human primates (NHPs) and rats. The average signal duration per file was 24 hours.
SZR30a prototype performance was assessed by comparing automated analysis (using the module default settings) with manual marking in terms of:
- Sensitivity: ability of the analyzer to detect seizure
- False positive per hour: number of false detections per hour of EEG signal
The table below shows the results for the Rat and the NHP species.
Artifacts in the rat EEG signals and electromyogram crosstalk relative to chewing in the NHPs are the main sources of false detections. Nevertheless, 100% detection sensitivity was reached for both species when selecting optimal settings for each individual.
We are currently investigating further optimization techniques in order to improve detection sensitivity and minimize the number of false detections.
To provide a complete solution, we are simultaneously working on spike detection. Performance was assessed on the same EEG signals database for rats only. We achieved an average sensitivity of 88.8%. The user can define exclusion parameters based on spike amplitude, duration and baseline amplitude to refine the analysis.
Once spike detection is over, the train detection allows to group them into spike train. This detection relies on the parameters described above and on additional parameters specific to train: minimum train duration, the latency between spikes and the number of consecutive spikes.
SPS 2012, don't miss:
► Presentation: "Epileptic Seizures Characterization and Automated Detection in Telemetered Nonhuman Primates" by Simon Authier et al., from CiToxLAB
- Tuesday, Oct. 2nd - 2.00 to 2.15 pm - Room 150-152
► Poster #108