Authors :
Presenting Author: Peter Schwab, MD – University of Washington
Erin Conrad, MD – University of Pennsylvania; Ryan Gallagher, MD Candidate – University of Pennsylvania; Nicole Hartmann, DO, MBS – University of Pennsylvania; Joshua Larocque, MD, PhD – University of Pennsylvania
Rationale:
Staged surgical treatment for drug-resistant epilepsy is costly and outcomes are variable. Patterns of interictal epileptiform discharges (IEDs) have been identified as a useful biomarker to predict favorable surgical candidacy. The utility of these data is limited by the time and expertise constraints inherent in direct EEG reading. Commercially available automated IED detection software allows for quantitative characterization of these phenomena in pre-operative scalp EEG data. Here we studied the accuracy of supervised automated IED detections against expert reviewers and the concordance with clinician-defined seizure onset zone localization.
Methods:
We used a previously published supervised spike detection approach (Epilepsia 2022; 63:1064-1073) with Persyst 13 (Journal of Clinical Neurophysiology 2021; 38(5): 439-447) to identify IEDs in scalp EEG data from pre-operative EMU admissions in a group of 38 sequential patients who ultimately underwent staged surgical treatment. Three epileptologists reviewed automatically detected spike morphologies and scored them as true IEDs or artifactual. We quantified the agreement between automatically detected IED patterns and clinician-defined spike patterns and between automatically detected IED laterality and final seizure onset zone localization.
Results:
34/38 subjects had IEDs detected with Persyst. Mean reviewer inter-rater reliability (kappa statistic) in determining which detected IEDs were true IEDs was 0.45 (IQR 0.16-0.76). 25/34 subjects with IEDs detected were deemed to have at least one population of true IEDs. The laterality (left, right or bilateral) of these populations of detected, reviewed IEDs were found to agree with the IED laterality (left, right, or bilateral) determined by clinicians for 68% of subjects (17/25), and with final post-implant laterality in 44% of subjects (11/25).
Conclusions:
A supervised spike detection algorithm using commercially available software can accurately lateralize scalp EEG IEDs. This study is significant because a supervised approach to quantitative IED detection may help expedite review of scalp EEG for surgical planning.Funding: NIH 5-U24-NS 063930 The International Epilepsy Electrophysiology Database