Abstracts

An Automated Algorithm for MEG Language Mapping Using Equivalent Current Dipole and Comparison with Other Methods

Abstract number : 2.028
Submission category : 3. Neurophysiology / 3D. MEG
Year : 2022
Submission ID : 2204135
Source : www.aesnet.org
Presentation date : 12/4/2022 12:00:00 PM
Published date : Nov 22, 2022, 05:23 AM

Authors :
Abbas Babajani-Feremi, PhD – Dell Medical School, University of Texas at Austin, Austin, TX, USA; Haatef Pourmotabbed, MSC – Neurology – Dell Medical School, University of Texas at Austin, Austin, TX, USA; William Schraegle, PhD – Neurology – Dell Medical School, University of Texas at Austin, Austin, TX, USA; Clifford Calley, MD – Neurology – Dell Medical School, University of Texas at Austin, Austin, TX, USA; Dave Clarke, MD – Neurology – Dell Medical School, University of Texas at Austin, Austin, TX, USA; Andrew Papanicolaou, PhD – Pediatrics – University of Tennessee Health Science Center, Memphis, TN, USA

This abstract has been invited to present during the Neurophysiology platform session.

Rationale: Magnetoencephalography (MEG) is still underutilized in presurgical language mapping in epilepsy centers, though its utility and accuracy have been demonstrated.1 The single equivalent current dipole (sECD) is the best validated approach for MEG language lateralization,2 however, it remains difficult to implement reliably due to subjective judgements made by the user. To address this technical issue, we developed a completely objective and automated sECD algorithm (AsECDA) for language mapping. We then compared performances of AsECDA and three most popular methods: minimum norm estimation (MNE), dynamic statistical parametric maps (dSPM), and dynamic imaging of coherent sources (DICS) event-related desynchronization (ERD) and event-related synchronization (ERS) beamformer.

Methods: The AsECDA was first evaluated using synthetic MEG datasets by considering dipoles in the centroid of 246 cortical areas in the Brainnetome atlas and various signal-to-noise ratios (SNRs). Then MEG data collected in 21 patients with epilepsy (23 ± 17 years; 11 males; 18 right- and 3 left‐handed) during a word recognition task2 in two repeated sessions were used to compare performance of four aforementioned methods.

Results: Results of synthetic MEG datasets revealed a good localization accuracy for the AsECDA in both superficial and deep sources, and the average localization error in a typical noise level (SNR = 5) was less than 2 mm (1.6 ± 0.6 mm). Figure 1 shows language localization using the four methods in a representative patient. As depicted in Table 1, the AsECDA provided a strong and significant correlation between the calculated laterality index (LI) of two sessions across all patients (Cor = 0.80; P < 0.00002). This correlation for MNE (Cor = 0.71) and dSPM (Cor = 0.64) were smaller than that for AsECDA. In addition, DICS-ERD in alpha and low beta bands provided fair correlations (Cor = 0.64 and 0.48, respectively) between the calculated LIs of the two sessions, and this correlation was nonsignificant (P > 0.1) for DICS-ERD in high beta and gamma bands and for DICS-ERS in all frequency bands. The DICS-ERD in alpha and low beta bands, MNE, and dSPM inclined toward providing a biased result toward symmetric language representation, and > 50% of the sessions were assigned by these methods to right laterality or bilateral in contrast with previous studies of language lateralization in epilepsy patients.3 The AsECDA provided the best results compared to other methods regarding lateralization and > 62% of the sessions were assigned by AsECDA to left laterality.

Conclusions: The AsECDA provided superior results compared to MNE, dSPM, and DICS beamformer regarding the test-retest reliability and the expected left lateralization and correct localization of the receptive language cortex._x000D_  _x000D_ References:
1. Bowyer et al. J Clin. Neurophysiology. 2020;37:554-563.
2. Papanicolaou et al. J Neurosurgery. 2004;100(5):867-876.
3. Gaillard et al. Neurology. 2007;69(18):1761-1771.

Funding: None
Neurophysiology