Abstracts

Comparison of 11 Automatic Detectors of High Frequency Oscillations in Scalp and Intracranial EEG

Abstract number : 3.172
Submission category : 3. Neurophysiology / 3G. Computational Analysis & Modeling of EEG
Year : 2023
Submission ID : 1010
Source : www.aesnet.org
Presentation date : 12/4/2023 12:00:00 AM
Published date :

Authors :
Presenting Author: Margarita Maltseva, MD – Alberta Children's Hospital Research Institute & Hotchkiss Brain Institute, Cumming School of Medicine, University of Calgary, AB, Canada

Daniel Lachner-Piza, PhD – Alberta Children's Hospital Research Institute & Hotchkiss Brain Institute, Cumming School of Medicine, University of Calgary, AB, Canada; Minette Krisel Manalo, MD – Neuroscience – Alberta Children's Hospital Research Institute & Hotchkiss Brain Institute, Cumming School of Medicine, University of Calgary, AB, Canada; Julia Jacobs-LeVan, MD – Alberta Children's Hospital Research Institute & Hotchkiss Brain Institute, Cumming School of Medicine, University of Calgary, AB, Canada

Rationale:
Innovative research on epilepsy has focused on developing new biomarkers for epilepsy including high-frequency oscillations (HFOs). Pathological HFOs in the range between 80-500Hz have been investigated mainly in intracranial electroencephalography (EEG). HFO research accelerated when pathological interictal HFOs were also discovered in non-invasive scalp EEG. However, researchers still face the challenges of highly time consuming visual marking. Therefore, automatic detectors are crucial for systematic study of HFOs and clinical application. Especially in scalp EEG, precision and reliability of automatic HFO detectors is unclear and exciting detectors largely vary in used methodology. This study focuses on comparing detection result from existing detectors in both scalp and intracranial EEG using the identification of seizure onset areas as gold standard.



Methods:
Files from 15 patients with both intracranial EEG and scalp EEG, were selected retrospectively from our EEG database. At least 30 minutes of continuous EEG recording from sleep stage II, with low level of artefacts and no electroclinical seizures were analyzed. A total of 11 different published automatic detectors were included in this comparison. All detectors were run with their default parameters on both scalp and intracranial EEG.

Results:
All HFO detectors identified the seizure onset zone (SOZ) with higher significance than a random classifier (Figure 1 and 2, Column B). However, only one detector reached an area under receiver operating characteristic curve (AUC) of at least 0.7, indicating acceptable validity. In intracranial EEG, agreement between detectors on spikes co-occurring with spikes (spike-HFO), based on interpretation of AUC, was higher compared with HFO only occurrence rate per EEG channel (Figure 1, Column B). In scalp EEG, agreement between detectors on HFO only occurrence rate was higher compared with spike-HFO (Figure 2, Column B). The average value of Spearman's rho and overall agreement between detectors were higher detecting HFO occurrence rate in scalp EEG (Figure 1, Column A). Detectors did not necessarily perform better in the recording methods they were designed for (scalp and intracranial respectively).

Conclusions:
To the best of our knowledge, this is the first study comparing 11 automatic HFO detectors while using them on the same database. Variability of HFO detections between different detectors is high and mirrors the variability that clinicians face during visual marking. Surprisingly, detections of scalp HFOs showed comparable results with detections in intracranial. While the agreement on identifying individual HFO events was low between detectors, there was a high agreement in identifying the areas most active in generating HFO. This is promising when aiming to identify epileptic areas. More detailed analysis of HFO characteristics and resulting detector performance will be necessary to better understand why distinct detectors show higher performance in some patients.

Funding:
This work was supported by the Canadian Institutes of Health Research grant number 480576.

Neurophysiology