Real-time detection of absence seizures

Objective: Absence seizures are characterized by changes in attention, potentially leading to hazardous situations. We developed a real-time seizure detection algorithm for online detection of absences. Our aim is to integrate this algorithm into an assessment application, enabling the measurement of attention during absences.

Methods: Our algorithm uses a continuous wavelet transform of single-channel EEG data. We tested the algorithm offline on 22 continuous 24-hour EEG recordings of pediatric patients with absences. We externally validated our algorithm with 49 routine EEGs and twelve ambulatory recordings. To quantify the algorithm's performance, we determined sensitivity, false positive rate, time to first detection and F1 score.

Results: In our test dataset, we obtained an average sensitivity of 97.9 %, a false positive rate of 1.20/h and an F1 score of 0.82. Except for one patient, the median time until detection was <2 s. In our validation set, an average sensitivity of 95 % and an F1 score of 0.88 was reached. The average false positive rate was 5.6/h and 4.5/h for the routine and ambulatory recordings, respectively. The median time until first detection was 1.1 s.

Conclusions: Our algorithm demonstrates fast detection of absence seizures with a high sensitivity, making it suitable for integration in a computerized reaction time task. The high false positive rate indicates the importance of a careful review of the results. Significance: The algorithm has the potential to play a useful role in the advancement of clinical and research applications aimed at studying transient impaired attention during absences.