Characterization of Physiological High-frequency Oscillations Obtained from Chronic Intracranial EEG in Children Using Deep Learning
Abstract number :
1.192
Submission category :
3. Neurophysiology / 3G. Computational Analysis & Modeling of EEG
Year :
2022
Submission ID :
2204738
Source :
www.aesnet.org
Presentation date :
12/3/2022 12:00:00 PM
Published date :
Nov 22, 2022, 05:26 AM
Authors :
Jacqueline Ngo, MD – UCLA; Yipeng Zhang, MS – Electrical and Computer Engineering – UCLA; Hoyoung Chung, BS – Electrical and Computer Engineering – UCLA; Tonmoy Monsoor, MS – Electrical and Computer Engineering – UCLA; Shaun Hussain, MD, MS – Pediatrics – UCLA; Joyce Matsumoto, MD – Pediatrics – UCLA; Patricia Walshaw, PhD – Psychiatry – UCLA; Aria Fallah, MD, MS – Neurosurgery – UCLA; Myung Shin Sim, PhD – Medicine – UCLA; Raman Sankar, MD, PhD – Pediatrics and Neurology – UCLA; Richard Staba, PhD – Neurology – UCLA; Jerome Engel, MD, PhD – Neurology – UCLA; William Speier, PhD – Radiological Sciences and Bioengineering – UCLA; Vwani Roychowdhury, PhD – Electrical and Computer Engineering – UCLA; Hiroki Nariai, MD, PhD, MS – Assistant Professor, Pediatrics, UCLA
Rationale: Intracranially-recorded interictal high-frequency oscillations (HFOs) have been proposed as a promising spatial biomarker of the epileptogenic zone. However, HFOs can also be recorded in the healthy brain regions (physiological HFOs), which complicates the interpretation of HFOs. The present study aimed to characterize salient features of physiological HFOs using deep learning.
Methods: We constructed a deep learning (DL)-based algorithm using HFOs from chronic intracranial electroencephalogram (iEEG) data via subdural grids from children with medication-resistant neocortical epilepsy, with confirmed functional cortical mapping results and post-operative seizure outcomes. Time-series EEG data were transformed into imaging inputs, and a deep learning (DL) algorithm was trained with functional cortical mapping results as DL labels for a weak-supervised approach. Morphological characteristics of HFOs obtained from the eloquent cortex without pathological findings (spikes and seizure onset zones) were distilled and interpreted through the DL model to represent phyiological HFOs (defined as ecHFOs).
Results: A total of 63,379 interictal intracranially-recorded HFOs were analyzed from 18 patients (age range 3–20 years). The ecHFOs had lower amplitude throughout the frequency band (80-500 Hz) around the HFO onset and also had a lower signal amplitude in the low frequency band throughout the time window than non-ecHFOs, resembling a bell-shaped template in the time-frequency map (Figure 1A/B). Such morphological characteristics were confirmed to influence model prediction via perturbation analyses (Figure 1C/D). We noted that 22.9% (4547/19816) of the ecHFOs were HFOs with spikes, and 70.4% (30647/43563) of non-ecHFOs were HFOs with spikes. We plotted the histogram of the amplitude, length, and max frequency of both ecHFO and non-ecHFO, respectively (Figure 2). The ecHFOs exhibited a smaller amplitude (p-value < 0.01), a lower max frequency (p-value < 0.01) and a trend of shorter HFO length (p-value = 0.07) than non-ecHFO. Using the resection ratio (removed HFOs/detected HFOs) of non-ecHFOs, the prediction of post-operative seizure outcomes improved compared to using the uncorrected HFO resection ratio (area under the ROC curve of 0.82, increased from 0.76).
Neurophysiology