Career Center AES Connect

Abstracts

A Comparison of Seizure Onset Zone Localisation Algorithms with Intracranial EEG: Evaluating Methodological Variations and Targeted Features

Abstract number : 2.201
Submission category : 3. Neurophysiology / 3G. Computational Analysis & Modeling of EEG
Year : 2024
Submission ID : 269
Source : www.aesnet.org
Presentation date : 12/8/2024 12:00:00 AM
Published date :

Authors :
Sarah J Gascoigne, BSc, MSc – Newcastle University
Presenting Author: Manel Vila-Vidal, PhD – BrainFocus & NCODE lab

Nathan Evans, Mphys – Newcastle University
Adrià Tauste, PhD – NCODE lab & BrainFocus
Yujiang Wang, PhD – Newcastle University

Rationale: Automatic seizure onset zone localisation algorithms are being used more frequently in research and in clinical settings. Each algorithm is designed to highlight changes in different signal features associated with seizure onset. This work is the first to investigate how methodological differences in SOZ algorithms can result in different onset localisations. Further, we demonstrate that a blanket application of onset localisation algorithms assuming that the performance will be just as good across all subjects and seizure types, may not be the most appropriate approach.

Methods: We analyzed ictal intracranial EEG (icEEG) recordings from 100 seizures in 16 patients with drug-resistant epilepsy, using data from the SWEZ-ETHZ public database. Through our analysis, we identified a series of critical decision points that must be considered when designing or selecting a seizure onset localisation algorithm. These points were demonstrated using three distinct algorithms that capture different but complementary seizure onset features: Imprint (Gascoigne et al., 2024), Epileptogenicity Index (EI, Bartolomei et al., 2008), and Seizure Activation Map (SAM, Vila-Vidal et al., 2020).


Results: Our independent application of these algorithms on the above dataset revealed significant impacts of each decision point on the resultant seizure onset localisations. Pairwise comparisons of the algorithms showed low agreement, with 27-60% of seizures classified as minimally or non-overlapping. We assessed the methodological differences at each key decision point, comparing the resultant onsets given different choices. For instance, some decision points resulted in no onset being found, such as 26% of seizures having no SAM onset due to an entropy threshold limiting the proportion of channels labeled as onset. Conversely, 13.1% of Imprint onsets captured decreases in activity, a feature not captured by other algorithms. Finally, visually similar seizure onsets can have different localisations using EI if the rate of increase in high-frequency activity does not exceed the threshold in one or more channels.

Conclusions: This work highlights the heterogeneity of SOZ signal features exhibited within and across individuals, suggesting that finding a gold standard for SOZ localisation which can be applied across or within subjects is nowadays a challenging task.Our results support previous findings that seizure features vary within individuals, as onset patterns differed for all subjects across all algorithms. Finally, we provide a recommendation that all decision points must be considered when selecting or designing a seizure onset localisation algorithm.

Funding: SJG, NE, YW: The Engineering and Physical Sciences Research Council, Centre for Doctoral Training in Cloud Computing for Big Data (grant number EP/L015358/1), Wellcome Trust (208940/Z/17/Z), UKRI Future Leaders Fellowship (MR/V026569/1) and (MR/T04294X/1). ATC, MVV: Spanish Ministry of Science, Innovation, and Universities (MCIU/AEI /10.13039/501100011033), Grant/Award Number ID2020-119072RA-I00/AEI/10.13039/501100011033 and Grant/Award Number TQ2022-012679

Neurophysiology