Connectome spectrum electromagnetic tomography: application in focal epilepsy source localization

1665 

Abstract Submission 

Nino Herve¹, Joan Rué-Queralt², Yasser Alemán-Gómez¹, Jonathan Wirsich³, Bernd Vorderwülbecke⁴, Laurent Spinelli³, Margitta Seeck³, Serge Vulliemoz³, Patric Hagmann¹

¹Department of Radiology, Lausanne University Hospital and University of Lausanne (CHUV-UNIL), Lausanne, Switzerland, ²Center for Imaging, École Polytechnique Fédérale de Lausanne, Lausanne, Switzerland, ³EEG and Epilepsy Unit, University Hospitals and Faculty of Medicine University of Geneva, Geneva, Switzerland, ⁴Epilepsy Center Berlin-Brandenburg, Department of Neurology, Charité-Universitätsmedizin Berlin, Berlin, Germany

Nino Herve  
Department of Radiology, Lausanne University Hospital and University of Lausanne (CHUV-UNIL)
Lausanne, Switzerland

Joan Rué-Queralt  
Center for Imaging, École Polytechnique Fédérale de Lausanne
Lausanne, Switzerland

Yasser Alemán-Gómez  
Department of Radiology, Lausanne University Hospital and University of Lausanne (CHUV-UNIL)
Lausanne, Switzerland

Jonathan Wirsich  
EEG and Epilepsy Unit, University Hospitals and Faculty of Medicine University of Geneva
Geneva, Switzerland

Bernd Vorderwülbecke  
Epilepsy Center Berlin-Brandenburg, Department of Neurology, Charité-Universitätsmedizin Berlin
Berlin, Germany

Laurent Spinelli  
EEG and Epilepsy Unit, University Hospitals and Faculty of Medicine University of Geneva
Geneva, Switzerland

Margitta Seeck  
EEG and Epilepsy Unit, University Hospitals and Faculty of Medicine University of Geneva
Geneva, Switzerland

Serge Vulliemoz  
EEG and Epilepsy Unit, University Hospitals and Faculty of Medicine University of Geneva
Geneva, Switzerland

Patric Hagmann  
Department of Radiology, Lausanne University Hospital and University of Lausanne (CHUV-UNIL)
Lausanne, Switzerland

To avoid invasive procedures, scalp-recorded EEG can be used to solve inverse problems for localizing focal epileptic sources. Connectome spectrum electromagnetic tomography (CSET) is an innovative EEG source reconstruction method that considers structural connections by imposing sparsity on the connectome spectrum (Rué-Queralt 2022). In this work, CSET was tested and compared to weighted minimum norm estimation (WMNE) using one patient from the dataset in Vorderwülbecke 2022. The patient achieved epilepsy-free status after the surgical removal of the epileptic source.

Multimodal processing pipeline is illustrated in Figure 1.
MRI: Pre/post-surgical structural 3T MRI were acquired with 0.7/1.0 mm slice thickness. T1-MPRAGE images were re-sampled to 1mm3 isotropic resolution through cubic interpolation using Connectome Mapper 3, and artifacts were removed using Cartool. Post-surgical MRI data was acquired three months after surgery and linearly co-registered to pre-surgical MRI using FSL5.0.
EEG: Pre-surgical 256-channel (EGI) EEG was recorded for around 20 minutes. Interictal discharges were identified visually by a trained specialist, high-pass filtered at 0.3-1 Hz and low-pass filtered at 30-100 Hz. Epochs were centered on the discharge peak before being averaged and downsampled to 250 Hz. The time point at 50% of the averaged discharge's rising phase is utilized for source reconstruction.
Forward model: Brain, skull and skin surfaces were extracted from pre-surgery T1 MRI using the watershed algorithm implemented in Freesurfer 7.2.0. Python MNE 1.5.0 library was used to construct the boundary element model, sampling cortical brain sources at ~5 mm intervals (~11000 cortical sources) and the computation of the lead field matrix.
Connectome: The source connectome was created from the probabilistic connectome atlas (Alemán-Gómez 2022). The patient's brain underwent the same parcellation as the atlas to associate each source to one of the 446 cortical regions. Each source inside a region is connected to all the sources in another region if the regions exhibit consistency in connectivity higher than 75%. We expanded the network by including geodesically close connections to accommodate grey matter interactions.
Inverse Problem: The task involves finding the brain activity distribution, denoted as x, that minimizes the optimization problem: argmin_x {1/2 ||b-Ax||²₂+||Fx||₁}. b represents EEG measurements, A is the lead field matrix, and F is the graph Fourier transform of the connectome. Fx is the connectome spectrum, and minimizing ||Fx||₁ enforces sparsity according to the user-defined parameter λ. The solution was obtained using proximal gradient descent implemented through Pyxu (Simeoni 2023).

Figure 2 shows reconstructed brain activity, where the last column projects peak activity into the voxel space, estimating spike origins in the right temporal pole for CSET and the left rectus and medial orbitofrontal gyri for WMNE. While WMNE showed a diffuse result, CSET had focused targets, aligning peak activity precisely with the surgically resected region where focal epilepsy was clinically localized.

In conclusion, the CSET method successfully identified the source of epilepsy within the resected area, a capability not achieved by WMNE. Incorporating the connectome into source reconstruction aims to discern a more fitting regularizer for the inverse problem, drawing insights from the previously unveiled structural-functional relationships (Glomb 2020). The dataset (including 44 more patients) allowed us to test reconstruction outcomes using actual data rather than simulations, providing a more realistic assessment of method efficacy.

Connectivity (eg. functional, effective, structural) ²

EEG/MEG Modeling and Analysis ¹

Electroencephaolography (EEG)

Epilepsy

Source Localization

STRUCTURAL MRI