Abstracts

Imaging Epileptic Sources from Scalp EEG HFOs Using a Spatial-Temporal-Spectral Imaging Framework

Abstract number : 1.166
Submission category : 3. Neurophysiology / 3G. Computational Analysis & Modeling of EEG
Year : 2023
Submission ID : 60
Source : www.aesnet.org
Presentation date : 12/2/2023 12:00:00 AM
Published date :

Authors :
Presenting Author: Xiyuan Jiang, MS – Carnegie Mellon University

Zhengxiang Cai, PhD – Carnegie Mellon University; Boney Joseph, M.B.B.S. – Mayo Clinic; Gregory Worrell, M.D., Ph.D. – Mayo Clinic; Bin He, Ph.D. – Carnegie Mellon University

Rationale:

Non-invasive EEG source imaging is a valuable technique to obtain information about epileptic sources, which aids in presurgical planning [1]. Electrophysiological source imaging (ESI) algorithms utilize various types of biomarkers present in EEG as input, such as seizures, high-frequency oscillations, and interictal spikes. It has been observed that the extent of the source is inversely proportional to its frequency; higher frequency sources are smaller in size and produce weaker EEG signals on the scalp. There is a clinical need to image sources generating both low and high-frequency biomarkers. We developed a spatial-temporal-spectral source imaging framework that aims to achieve two objectives: 1) to delineate the EEG 3D tensor signal considering space, time, and frequency, into distinct components, and 2) to estimate the location and spatial extent of sources that generate these signals, using a data-driven L1 optimization problem considering both source sparsity and source edge sparsity [2]. We investigate here the performance of the proposed spatial-temporal-spectral imaging (STSI) framework in a group of focal drug-resistant epilepsy (fDRE) patients, and compare it with clinical ground truth.



Methods:

The time-frequency representation tensor of the EEG signal is decomposed into its respective components. Specific components are selected as input for an L1 regularization problem, wherein the objective function is a weighted sum of source sparsity and source edge sparsity. The resulting solution provides the current density of epileptic sources responsible for generating biomarkers of specific frequencies. Fourteen fDRE patients (ILAE outcome: I) with 40 HFO riding spike events [3] were selected to evaluate the performance of the STSI framework. Patient-specific head models and surgical resection were generated based on preoperative and postoperative MRI scans. The evaluation metrics used include localization error (LE), spatial dispersion (SD), precision, recall, and specificity.



Results:

The STSI framework could accurately estimate the location and extent of biomarkers with different frequencies. The average statistics for STSI-high/STSI-low/sLORETA are as follows. LE: 4.38/6.01/7.29 mm, SD: 5.45/8.15/14.87 mm, precision: 0.61/0.44/0.39, recall: 0.66/0.77/0.76, specificity: 0.94/0.88/0.87. The results indicate that in HFO riding spike events, high-frequency biomarkers provide better localization ability than concurrent low-frequency biomarkers.



Conclusions: We introduced an EEG source imaging framework that allows for the objective imaging of the location and extent of biomarkers with different frequencies. To evaluate the effectiveness of our proposed method, we conducted tests using EEG HFO riding spike events from 14 focal drug-resistant epilepsy patients. Our results demonstrate the feasibility of imaging the location and extent of epileptic sources using EEG spectral contents and the merit of performing source imaging on HFOs compared with interictal spikes.

Ref: 
1. He, B. et al., Ann. Rev. Biomed. Eng., 20 (2018): 171-196.
2. Sohrabpour, A. et al., Nat. Commun., 11.1 (2020): 1-15.
3. Cai, Z. et al., PNAS, 118, no. 17 (2021): e2011130118.

Funding: NIH NS096761

Neurophysiology