Characterizing Interictal EEG Waveforms in Lennox-gastaut Syndrome Using Time-frequency Image-based Features
Abstract number :
3.122
Submission category :
2. Translational Research / 2C. Biomarkers
Year :
2022
Submission ID :
2204330
Source :
www.aesnet.org
Presentation date :
12/5/2022 12:00:00 PM
Published date :
Nov 22, 2022, 05:24 AM
Authors :
Derek Hu, Doctoral Candidate – University of California, Irvine; Mandeep Rana, MD – Neurology – Children's Hospital of Orange County; David Adams, MD – Neurology – Children's Hospital of Orange County; Linh Do, IMG/EFCMG – Children's Hospital of Orange County; Daniel Shrey, MD – Children's Hospital of Orange County; Shaun Hussain, MD – UCLA Health; Beth Lopour, PhD – Biomedical Engineering – University of California, Irvine
Rationale: Lennox-Gastaut Syndrome (LGS) is an epileptic encephalopathy in which the EEG is typically characterized by the presence of various interictal epileptiform waveforms such as generalized paroxysmal fast activity (GPFA) and slow spike-and-wave. While these EEG patterns are relevant for LGS diagnosis, the features of these waveforms vary considerably, and the definition of GPFA lacks consensus. Here, we characterize the temporal, spectral, and spatial properties of interictal waveform patterns in LGS using unsupervised machine learning.
Methods: A ten-minute clip of EEG during non-REM sleep was retrospectively collected for each of eight subjects with LGS (age range 17-208 months) and four approximately age-matched healthy controls (age range 17-201 months). For each subject, the EEG time-frequency spectrogram was calculated and treated as a time-frequency image (TFI). Concurrent time-frequency points with significantly high-power were marked as events of interest (EOI). EOIs across all subjects were combined; they were then separated into ten clusters using K-means, based on six features of the EOI: (1) width, (2) height, (3) area, (4) spatial spread (number of electrodes), (5) frequency with maximum power, and (6) mean power. The EOIs for LGS subjects and healthy controls were compared, focusing on TFI features and the number of events from each subject group in each cluster.
Results: Automated time-frequency analysis across twelve subjects resulted in 4684 detected EOIs, with 1538 EOIs from controls and 3146 from LGS subjects. The EOI features for LGS subjects had a significantly greater height, spread, area, peak frequency, and mean power compared to healthy controls (Mann-Whitney U test, p< 0.05, Bonferroni corrected). Following k-means clustering analysis, the majority of EOIs (392/435) in the four clusters with the largest TFI features were found to originate from LGS subjects. The clusters with the highest TFI feature values were visually concordant with epileptiform waveforms; EOIs that resembled GPFA had peak frequencies around 12-15 Hz with high width, height, area, mean power, and spatial spread. EOIs that resembled slow spike-and-wave had less high frequency activity (lower height) and were shorter in duration (width) compared to GPFA. In contrast, most control subject EOIs were grouped in the cluster with the smallest TFI features and resembled non-epileptiform events.
Translational Research