Abstract
ABSTRACTObjectiveThe study was aimed at developing an automatic system, based on complex network analysis and machine learning, to identify interictal network-based biomarkers in patients with drug-resistant focal epilepsy and no visible anatomical lesions candidate for surgery, able to support the localization of the Epileptogenic Zone (EZ) and to further disclose properties of the interictal epileptogenic network.Methods3 min of interictal SEEG signals, recorded in 18 patients with drug-resistant epilepsy, different EZ localization, negative MRI, were analysed. Patients were divided into seizure-free (SF) and non-seizure free (NSF) groups, according to their post-surgical outcome. After a first step of effective connectivity estimation, hubs were defined through the combination of nine graph theory-based indices of centrality. The values of centrality indices related to these hubs were used as input of an ensemble subspace discriminant classifier.ResultsThe proposed procedure was able to automatically localise the EZ with 98% sensitivity and 59% specificity for SF patients. Moreover, our results showed a clear difference between SF and NSF patients, mainly in terms of false positive rate (i.e., the percentage of NEZ leads classified as EZ), which resulted significantly higher in NSF patients. Lastly, the centrality indexes confirmed a different role of the Propagation Zone in NSF and SF groups.SignificanceResults pointed out that network centrality plays a key role in interictal epileptogenic network, even in case of the absence of anatomical alterations and SEEG epileptic abnormalities, and that the combination of connectivity, graph theory, and machine learning analysis can efficiently support interictal EZ localization. These findings also suggest that poorer post-surgical prognosis can be associated with larger connectivity alteration, with wider “hubs”, and with a different involvement of the PZ, thus making this approach a promising biomarker for surgical outcome.Impact statementThe correct localization of the epileptogenic zone is still an unsolved question, mainly based on visual and subjective analysis of electrophysiological recordings, and highly time-consuming due to the needing of ictal recording. This issue is even more critical in patients with negative MRI and extra-temporal EZ localization. The approach proposed in this study represents an innovative and effective tool to reveal interictal epileptogenic network abnormalities, able to support and improve the EZ presurgical identification and to capture differences between poor and good post-surgical outcome
Publisher
Cold Spring Harbor Laboratory