Auditory Functional Mapping. Validation by simultaneous depth and MEG recordings

Objectives Hemispheric dominance for language is one of the most important to predict both risks of language production and verbal memory impairment. Evoked activities recorded from depth electrodes in the auditory areas have been proved to be efficient to reveal dominant hemisphere [1] . Similar information can also be obtained for non-invasive recordings (EEG or MEG) [2] . However, reconstruction of a cortical brain area from surface recordings is subject to the use of “inverse problems” that are still under development. Methods We have investigated patients with depth electrodes from up to 219 contacts with contacts located in the auditory areas and with MagnetoEncephaloGraphy. Thanks to technical developments, recordings have been performed simultaneously [3] . We used verbal sounds/BA/and/PA/are known to induce different cortical responses depending on the hemispheric lateralization [1] . Activity from virtual electrodes have been reconstructed from MEG recording using beam forming approach. Coordinates of the virtual electrodes have been taken from the depth electrodes contacts measured from IRM and CT scans. Results Reconstructions of virtual electrodes have been compared with recordings from the depth electrode. They presented similar time courses in both auditory and non-auditory areas. However, responses obtained from virtual electrodes are more spread than the invasive ones. General shapes and latencies of the event were correctly reconstructed allowing to evaluate the cerebral dominance from MEG data. It was also demonstrated in this study that close results can be obtained without MRI information. Conclusion Inverse solutions are known to be subject to a “leakage” effect that tends to spread the solution over a larger space than found with the seeg recordings. However, this demonstrates the ability of the MEG to unravel the cortical activity on a non-invasive way. Furthermore, the signals were properly reconstructed especially regarding the timing.


Data analysis
Epilepsy surgery requires the functional mapping of brain functions that might be affected.Recordings of evoked potentials during pre-surgical exploration by intracerebral electrodes give a very accurate information.These activities can also be obtained from non-invasive measurements (EEG or MEG).However, reconstructing the activity of a given cortical region requires the use of methods and models that are still discussed.
In this study, we recorded simultaneously from intracerebral electrodes and MEG sensors the activities evoked at the level of the auditory cortex in response to verbal sounds /ba/ and /pa/, electrophysiological index of hemispherical lateralisation (Liegeois-Chauvel et al. 1999 [1] , Trebuchon et al. 2005 [2] ).We used a spatial filtering technique to obtain, from the MEG measurements, the temporal time course of the evoked activities at the level of virtual electrodes defined at the location of the contacts of the intracerebral electrodes -thus allowing the comparison between these reconstructed responses and those coming from the intracerebral measurements.
Reconstruction for MEG at virtual electrode H'3 similar to the SVD-sphere estimation Note the supplementary responses corresponding to the voice onset time on the left side.Spatial filter provides the evoked response estimated for each SEEG contact from the MEG recordings: virtual electrode Software: Anywave [11] for visualization and preprocessing.Fieldtrip [12] for preprocessing and source reconstruction.For comparison purpose a normalization based on the maximum amplitude for each modality have been performed.

SEEG vs MEG virtual electrode reconstructions
Women 43 years old suffering for a left frontal lobe epilepsy SEEG: 13 electrodes in the left frontal and temporal lobes. 1 electrode in the right frontal lobe.The reconstructions are given here for the two electrodes exploring the left auditory cortical areas (electrode H').
Simultaneous MEG/SEEG: detail protocol in Dubabarry et al., 2014 [9] and Badier et al., 2017 [10] MEG: 248 full magnetometer system (3600WH 4D Neuroimaging).Important information in case of epilepsy surgery for the prediction of the risks of language deficit but also of verbal memory Several types of tasks tested in MEG, based on differences of evoked responses between the left and right hemispheres at the temporo-parietal region of the cortex: listening and word recognition [3,4] ; verbal production [5,6,7] ; semantic decision [8] .
"Bapa" protocol: passive listening, not requiring the cooperation of patients and/or children

OC' H'
Frontal-temporal Source reconstruction (inverse solution) by spatial filtering (LCMV [15] ) Estimation of the gain matrix (MEG forward solution) from the calculation of the magnetic field from one single source to each of the MEG sensor Head model anatomical registration between MEG and MRI by fiducial points and head surface matching determination of the SEEG contact coordinates (Epitools [13] ) segmentation of the patient head for construction of a realistic model of the head [14] Introduction Methods

Results and discussion
The temporal responses of the virtual sources obtained by spatial filtering of MEG data are similar to those obtained by SEEG data both in the primary auditory cortex region and for more distant regions.
The responses obtained in SEEG remain more focused than those obtained in MEG, which confirms the phenomenon of "source leakage" which tends to show a less punctual localization than expected and is due to the properties of the gain matrix.
the different modalities Evoked responses for MEG-SVD-Sphere Spheres for dipole selection

─
Head model: template (Colin 27) Source model: 3D grid of dipole uniformly spaces of 10 mm.The dipole retains are the one contained on a sphere (one for each side) centered around Heschl gyrus (104 dipoles).The first component of the resulting singular value decomposition (SVD) of the 104 dipoles is taken as the reconstruction.Ba (0.62) ─ Pa (0.81) ─ Ba (0.71) ─ Pa (0.67) Sampling frequency 2034.51Hz SEEG : 219 channels (BrainAmp ® ).Sampling frequency 2500 Hz.MEG and SEEG are synchronized through the stimulation triggers.and /pa/ displayed in random order, N=250 each, ISI 0.8-1.2s, two filter coefficients estimated from the variance of all trials and the gain matrix optimal orientation of the sources calculated from a singular value decomposition normalization of the temporal course of the virtual electrode by pseudo Z transformation using the baseline [-200 to 0 ms] • • • The answers obtained in MEG are able to allow a functional mapping of the auditory cortex areas.It is possible to obtain close results without the contribution of the MRI of the patient (cf.supplementary analysis) a Aix-Marseille Université, Institute of Language, Communication and the Brain (Aix-en-Provence, France) b Aix-Marseille Université and INSERM, Institut de Neurosciences des Systèmes (Marseille, France) c AP-HM, La Timone, Neurophysiologie Clinique (Marseille, France) d AP-HM, La Timone, Neurochirurgie Fonctionnelle et Stéréotaxique (Marseille, France)