Automated seizure detection in an EMU setting: Are software packages ready for implementation?

Area of Science:

• Clinical Neurophysiology
• Medical Device Technology
• Epilepsy Diagnostics

Background:

• Automated seizure detection software aims to improve the efficiency and accuracy of seizure identification in electroencephalography (EEG).
• Evaluating the performance of commercial software packages is crucial for their effective implementation in clinical settings, such as epilepsy monitoring units (EMUs).

Purpose of the Study:

• To assess the reliability of automated seizure detection using three commercial software packages (Persyst, Encevis, BESA) combined with live observation.
• To compare the performance of these software packages against clinical physiologists' review in an EMU setting.

Main Methods:

• Retrospective analysis of 286 prolonged EEG records from individuals aged 16-86 years.
• Comparison of seizure detection sensitivity and false positive rates across Persyst, Encevis, and BESA software.
• Reference standard included clinical reports and validated software detections.

Main Results:

• Seizure detection sensitivity was high for all software (95-98% for detecting at least one seizure).
• Persyst demonstrated the highest sensitivity (93%) for recognizing all seizures, followed by Encevis (88%) and BESA (84%).
• Clinical physiologists achieved 100% sensitivity at record level and 98% at seizure level, with lower false positive rates than most software.

Conclusions:

• Automated seizure detection software, while sensitive, does not outperform experienced clinical physiologists.
• These tools can be valuable in EMUs when users understand their limitations.
• Persyst software exhibited the strongest performance among the evaluated automated detection systems.