Accuracy and spatial properties of distributed magnetic source imaging techniques in the investigation of focal epilepsy patients

Abstract

Source localization of interictal epileptiform discharges (IEDs) is clinically useful in the presurgical workup of epilepsy patients. We aimed to compare the performance of four different distributed magnetic source imaging (dMSI) approaches: Minimum norm estimate (MNE), dynamic statistical parametric mapping (dSPM), standardized low‐resolution electromagnetic tomography (sLORETA), and coherent maximum entropy on the mean (cMEM). We also evaluated whether a simple average of maps obtained from multiple inverse solutions (Ave) can improve localization accuracy. We analyzed dMSI of 206 IEDs derived from magnetoencephalography recordings in 28 focal epilepsy patients who had a well‐defined focus determined through intracranial EEG (iEEG), epileptogenic MRI lesions or surgical resection. dMSI accuracy and spatial properties were quantitatively estimated as: (a) distance from the epilepsy focus, (b) reproducibility, (c) spatial dispersion (SD), (d) map extension, and (e) effect of thresholding on map properties. Clinical performance was excellent for all methods (median distance from the focus MNE = 2.4 mm; sLORETA = 3.5 mm; cMEM = 3.5 mm; dSPM = 6.8 mm, Ave = 0 mm). Ave showed the lowest distance between the map maximum and epilepsy focus (Dmin lower than cMEM, MNE, and dSPM, p = .021, p = .008, p < .001, respectively). cMEM showed the best spatial features, with lowest SD outside the focus (SD lower than all other methods, p < .001 consistently) and high contrast between the generator and surrounding regions. The average map Ave provided the best localization accuracy, whereas cMEM exhibited the lowest amount of spurious distant activity. dMSI techniques have the potential to significantly improve identification of iEEG targets and to guide surgical planning, especially when multiple methods are combined.

Evaluation of a personalized functional nearinfra‐redoptical tomography workflow using maximum entropy on the mean

“…While fitting the data through relative entropy maximization, MEM has the unique ability to switch off parcels of the model considered as inactive using a hidden variable. In our previous studies in the context of EEG/MEG source imaging, we have demonstrated excellent accuracy of MEM and the ability to be sensitive to the spatial extent of the underlying generators (Chowdhury et al, 2013;Chowdhury et al, 2016;Grova et al, 2016;Hedrich et al, 2017;Heers et al, 2016;Pellegrino et al, 2020), before adapting this framework in the context of NIROT (Cai et al, 2021). As the most conventional inverse procedure considered in NIROT, MNE is a linear method (Hämäläinen & Ilmoniemi, 1994) using Tikhonov regularization to minimize the L2-norm.…”

Diffuse optical reconstructions of fNIRS data using Maximum Entropy on the Mean

“…3, Table .1 and supplementary section S2), we found that MNE overall reconstructed well the main generator but largely overestimated the size of the underlying generator. MEM was specifically developed, in the context of EEG/MEG source imaging, as a method able to recover the spatial extent of the underlying generators, which has been proved not to be the case for MNE-based approaches (Chowdhury et al, 2013(Chowdhury et al, , 2016Grova et al, 2016;Hedrich et al, 2017;Pellegrino et al, 2020). A recent review (Sohrabpour and He, 2021) in the context of EEG/MEG source imaging has also demonstrated that the Bayesian approach with sparsity constraints is required to accurately estimate the spatial extent.…”

“…The MEM framework was specifically designed and evaluated for its ability to recover spatially extended generators (Heers et al, 2016;Pellegrino et al, 2016;Chowdhury et al, 2016;Grova et al, 2016). We recently demonstrated its excellent performances when dealing with focal sources (Hedrich et al, 2017) and when applied on clinical epilepsy data (Chowdhury et al, 2018;Pellegrino et al, 2020). In addition to its unique ability to recover the spatial extent of the underlying generators, we also demonstrated MEM's excellent accuracy in low SNR conditions, with the ability to limit the influence of distant spurious sources (Chowdhury et al, 2016;Hedrich et al, 2017;Heers et al, 2016;Pellegrino et al, 2020;von Ellenrieder et al, 2016;Aydin et al, 2020).…”

“…Maximum Entropy on the Mean (MEM) framework was first proposed by (Amblard et al, 2004) and then applied and carefully evaluated by our group in the context of EEG/MEG source imaging (Grova et al, 2006; Chowdhury et al, 2013). The MEM framework was specifically designed and evaluated for its ability to recover spatially extended generators (Heers et al, 2016; Pellegrino et al, 2016; Chowdhury et al, 2016; Grova et al, 2016), whereas we recently demonstrated excellent performances when dealing with focal sources as well (Hedrich et al, 2017) and when applied on clinical epilepsy data (Chowdhury et al, 2018; Pellegrino et al, 2020).…”

Ant Colony System Optimization for Spatiotemporal Modelling of Combined EEG and MEG Data

“…Finally, it makes use of non-convex sparsity priors and joint inference of source estimates to obtain accurate source amplitudes. Various applications for MEG/EEG source reconstruction have been applied in the clinical setting for detection of epileptic spikes [ 28 ], identification of seizure onset zone [ 29 ] and presurgical workup of epilepsy patients [ 30 , 31 , 32 ].…”