Deep learning-based early detection of absence seizures in children
Résumé
Background: Childhood Absence epilepsy is a common generalized epileptic syndrome in children. The seizures involve momentary lapses in consciousness, aligning with generalized spike-wave discharges on EEG. This study proposes an AI-based algorithm for early detection of absence seizure onset, allowing for application of sensory stimulation, such as acoustic stimulation, prone to abort the seizure. Method: We propose a deep learning-based model designed for early detection of the onset of absence seizures in children. The use of deep learning algorithms offers a promising solution to this problem by leveraging their ability to analyze complex patterns. The model was evaluated under two configurations. Clinical configuration to assess feasibility of an accurate detector of the onset seizures, and a wearable device configuration intended to implement the model on a portable closed-loop stimulator. Results: The performance analysis, in term of accuracy and time delay, assessed on a clinical EEG database of 117 patients with confirmed childhood absence epilepsy, are promising: sensitivity of 0.859, precision of 0.819, F1-score of 0.837, and a mean time delay of 0.522 s. Furthermore, the algorithm performance evaluated using reduced number of electrodes, as required for a wearable device, is still stable with a sensitivity = 0.837, precision = 0.808, F1-score = 0.820 and detection delays around 0.5 s. Conclusion: The performance of the proposed method on clinical configuration demonstrates the feasibility of a robust and universal detector of the onset of absence seizures in children. In addition, the consistency of results when only two bipolar EEG channels are utilized makes the pipeline suitable to be embedded in a wearable stimulator.