Abstract
Epilepsy affects over 50 million people worldwide, with approximately 30% experiencing drug-resistant forms that may require surgical intervention. Accurate localisation of the epileptogenic zone (EZ) is crucial for effective treatment, but how best to use intracranial EEG data to delineate the EZ remains unclear. Previous studies have used the directionality of neural activities across the brain to investigate seizure dynamics and localise the EZ. However, the different connectivity measures used across studies have often provided inconsistent insights about the direction and the localisation power of signal flow as a biomarker for EZ localisation. In a data-driven approach, this study employs a large set of 13 distinct directed connectivity measures to evaluate neural activity flow in and out the seizure onset zone (SOZ) during interictal and ictal periods. These measures test the hypotheses of "sink SOZ" (SOZ dominantly receiving neural activities during interictal periods) and "source SOZ" (SOZ dominantly transmitting activities during ictal periods). While the results were different across connectivity measures, several measures consistently supported higher connectivity directed towards the SOZ in interictal periods and higher connectivity directed away during ictal periods. Comparing six distinct metrics of node behaviour in the network, we found that SOZ separates itself from the rest of the network, allowing for the metric of "eccentricity" to localise the SOZ more accurately than any other metrics including "in strength" and "out strength." This introduced a novel biomarker for localising the SOZ, leveraging the discriminative power of directed connectivity measures in an explainable machine learning pipeline. By using a comprehensive, objective, and data-driven approach, this study addresses previously unresolved questions on the direction of neural activities in seizure organisation and sheds light on dynamics of interictal and ictal activity in focal epilepsy.