Abstracts

Rationale:
Seizure detection in human EEG recordings still remains an open and challenging task, despite numerous methods suggested since 1970s. The complexity of this task is exacerbated by the diversity of epilepsies and seizure types and difficulty in obtaining sufficiently clean signals in non-invasive EEG recordings. Improving automated seizure detection in generalized epilepsies may seem an easier task given that the widespread epileptic activity should be more reliably observed even in the presence of significant artefacts. However, it continues to be difficult for fully automatic algorithms to reliably detect seizures and other epileptiform activity without generating high rates of false positives.
Method:
We have developed a convenient semi-automatic interactive software (ASSYST). It has previously been used for reliable detection of EEG seizures in rodent models of acquired and genetic epilepsy. ASSYT uses an advanced time-frequency analysis that detects EEG episodes with excessive activity within a specific frequency band. In this report, we expanded its use and validate its ability to detect generalized spike and wave activity within a pool of independently recorded human EEG data. The software had modified parameters for this human population but was otherwise unchanged from that previously used in rodents. For these human recordings we applied the algorithm to find the spike component of the absence seizures, which in scalp recordings had a frequency within 15-22 Hz. 22 prolonged EEG recordings were obtained from 10 patients with generalized epilepsies and absence seizures, each lasting 24 hours. In individuals with repeat records there was a minimum of 7 days between studies. In addition, 2 EEG recordings with a limited number of electrodes organized in a C-shaped ear electrode grid obtained from 2 patients.
Results:
Both scalp and ear recordings contained expressed absence seizures (spike-wave discharges, SWD), as well as shorter interictal generalized epileptic discharges (GED). In 18 scalp recordings and 2 ear recordings the SWDs contained expressed spikes. In these recordings, the algorithm, tuned for 15-22 Hz band, detected 98% of SWDs. It was possible also to detect most of the GEDs. In 4 scalp EEG records the spike component was not expressed, however, it was still possible to detect 95% SWDs using a lower frequency band 2-6 Hz. The processing time with ASSYST depended on the number of seizures and artefacts in the recording and varied from6 to 96 min per 24 hours of EEG, or 30.3 min on average.This compared favourably with the time taken for an experienced researcher spending up to 2 to 3 hours to manually review, assess and annotate seizure activity in each 24 hour recording (giving a time reduction of 4 to 6 fold).
Conclusion:
Our ASSYST seizure detection tool provides high sensitivity, with acceptable specificity, for long and short-term EEG recordings from absence epilepsy patients. This has the potential to improve the efficiency and repeatability of clinical assessment and research. The successful detection of absence seizures recorded with the C-shaped ear electrode grid could enable a practical, self-contained wearable ambulatory EEG device.
Funding:
:
Funding:
to assist this research program was provided by Kaoskey Pty Ltd, Sydney, Australia