Imaging Spatiotemporally Distributed Epileptiform Sources from MEG Measurements
Abstract number :
2.034
Submission category :
3. Neurophysiology / 3D. MEG
Year :
2021
Submission ID :
1826015
Source :
www.aesnet.org
Presentation date :
12/5/2021 12:00:00 PM
Published date :
Nov 22, 2021, 06:51 AM
Authors :
Xiyuan Jiang, M.S. - Carnegie Mellon University; Abbas Sohrabpour - Carnegie Mellon University; Shuai Ye - Carnegie Mellon University; Vasileios Kokkinos - Massachusetts General Hospital; Alexandra Urban - University of Pittsburgh Medical School; Mark Richardson - Massachusetts General Hospital; Anto Bagić - University of Pittsburgh Medical School; Bin He - Carnegie Mellon University
Rationale: Non-invasive MEG/EEG source imaging provides valuable information about the epileptogenic brain areas to aid presurgical planning1-2. Our recently developed algorithm, FAST-IRES, objectively estimates extended sources and their time-course as it evolves3. Despite its success in imaging epileptogenic sources from EEG recordings, its application to other modalities such as MEG has not been attempted yet. In this work, through simulations and patient data analysis, we aim to assess the performance of FAST-IRES on imaging spatiotemporally distributed epileptogenic brain sources from MEG.
References:
1. He, B. et al., Ann. Rev. Biomed. Eng., 20 (2018): 171-196.
2. Bagić, A., Clin. Neurophysiol., 127.1 (2016): 60-66.
3. Sohrabpour, A. et al., Nat. Commun., 11.1 (2020): 1-15.
Methods: The default FreeSurfer cortex model was used in the simulations. Sources with varying extents of 10 – 40 mm were randomly selected at 500 locations on the cortex within a 60 mm range of the 102 magnetometer sensors. A spike-like signal was used to simulate the MEG temporal profiles and white Gaussian noise was added with SNR = 5, 10, 20 dB. Additionally, 10-min 306-channel MEG recordings of 5 focal drug-resistant epilepsy (fDRE) patients were obtained to test the performance of FAST-IRES in imaging epileptiform sources from interictal spikes. In data analysis, only the magnetometer data was used. The FAST-IRES algorithm was then applied on the averaged spikes. Two conventional methods, the linearly constrained minimum variance (LCMV) and standardized low-resolution brain electromagnetic tomography (sLORETA), were also applied after thresholding the solution. The evaluation metrics include localization error (LE), spatial dispersion (SD), and geometric mean (Geomean) of precision and recall.
Results: In simulation, FAST-IRES outperformed conventional methods in terms of LE, SD and geomean of precision and recall, under all SNR conditions (Fig. 1). We also found that FAST-IRES accurately estimates the extent of underlying sources. In 5 patients with fDRE with intracranial EEG recordings, the LE was compared with respect to seizure-onset-zone (SOZ) electrodes identified in the clinical report, and precision/recall was compared with respect to the SOZ. The average LE for LCMV, sLORETA and FAST-IRES were 21.55, 23.79, 11.97 (mm), and the average geomean of precision and recall were 0.08, 0.07 and 0.13, respectively. The difference was statistically significant between LE of sLORETA and FAST-IRES (p < 0.05, Wilcoxon rank-sum test). Fig. 2 shows the LE and Geomean of FAST-IRES analysis for the patients studied.
Neurophysiology