Towards the Automatic Localization of the Irritative Zone Through Magnetic Source Imaging

Abstract

The present work aims at validating a Bayesian multi-dipole modeling algorithm (SESAME) in the clinical scenario consisting of localizing the generators of single interictal epileptiform discharges from resting state magnetoencephalographic recordings. We use the results of Equivalent Current Dipole fitting, performed by an expert user, as a benchmark, and compare the results of SESAME with those of two widely used source localization methods, RAP-MUSIC and wMNE. In addition, we investigate the relation between post-surgical outcome and concordance of the surgical plan with the cerebral lobes singled out by the methods. Unlike dipole fitting, the tested algorithms do not rely on any subjective channel selection and thus contribute towards making source localization more unbiased and automatic. We show that the two dipolar methods, SESAME and RAP-MUSIC, generally agree with dipole fitting in terms of identified cerebral lobes and that the results of the former are closer to the fitted equivalent current dipoles than those of the latter. In addition, for all the tested methods and particularly for SESAME, concordance with surgical plan is a good predictor of seizure freedom while discordance is not a good predictor of poor post-surgical outcome. The results suggest that the dipolar methods, especially SESAME, represent a reliable and more objective alternative to manual dipole fitting for clinical applications in the field of epilepsy surgery.

An in-vivo validation of ESI methods with focal sources

“…First of all, the same dataset used here was used in [25], where an exemplar analysis with three ESI methods (MNE, ELOR and DSPM) was performed. Here, we considerably extended the comparison to include also more recent methods: noteworthy, we observed that recent methods such as SSM and MxNE do outperform older ones, hereby confirming recent results [19, 15]. In addition, we studied the impact of regularization and depth weighting parameters more in detail, highlighting similarities and differences between different methods.…”

“…Another study that relates quite directly to our current study is that in [19], where the authors compared retrospectively SSM, RAP and wMNE with the results of an ECD analysis on epileptic subjects; indeed, also in this case the reference source is a point source, even though in this case it is the product of a former analysis and not a true source; anyway, the authors find MNE to be the least accurate and SSM the most accurate.…”

“…One is to renounce the complexity of experimental data and use synthetic data to assess the reconstruction error of ESI [11,12,13,14,15,16]; this approach can result in reasonable comparisons between different methods but can hardly be used to give realistic estimates of localization accuracy in experimental scenarios, as the data generation process, particularly the forward model, is necessarily simplified. The second possibility, used in a growing number of studies, is to evaluate the accuracy of localization through post-surgical outcomes in epileptic patients [16,17,18,19,20]; this approach overcomes the limitations of synthetic data but has its own drawbacks, including the fact that resolution is limited to the size of the resected area. In this study we overcome the limitations of both approaches by exploiting a recently published EEG dataset of scalp recordings for which the ground truth is known.…”

Clinical Validation of the Champagne Algorithm for Epilepsy Spike Localization

“…In recent years, similar Bayesian iterative approaches have been applied to this problem in order to develop semi-automated algorithms for clinical use. For example, an automated multidipole iterative Monte Carlo approach called “SESAME” was evaluated by Luria et al ( 2020 ) and was concordant with standard ECD fitting done by experienced staff. Pellegrino et al ( 2018 ) applied the coherent Maximum Entropy of the Mean (cMEM) to yield distributed magnetic source images (dMSI) that provided improved accuracy compared to ECD, also having the advantage of user-independent results; as dMSI results provide spatial and temporal voxelwise information, they can offer information about anatomic extent of source beyond that available from ECD.…”

Clinical Validation of the Champagne Algorithm for Evoked Response Source Localization in Magnetoencephalography

“…These have included, for example, the application of the coherent Maximum Entropy of the Mean (cMEM) approach as described in Chowdhury et al (2013) to the localization of IEDs, a method termed distributed magnetic source imaging or dMSI, and found to be successful in comparison to ECD fitting by Pelligrino et al (2018). Recently, the Bayesian multidipole iterative Monte Carlo approach "SESAME" described by Sommariva and Sorrentino (2014) was evaluated in comparison to standard ECD methods by Luria et al (2020) for the localization of IEDs, likewise showing excellent performance. These methods were also noted to be advantageous because results were more operator-independent.…”