University of Twente Student Theses


A comparison of spherical and realistic head models for the detection of epileptiform activity in the temporal lobe with MEG

Wezel, F.A. van (2022) A comparison of spherical and realistic head models for the detection of epileptiform activity in the temporal lobe with MEG.

[img] PDF
Abstract:Epilepsy is a disease that affects around 50 million people worldwide. It causes seizures that can cause stares, unregulated muscle contractions, collapsing, tingling, or stiffness. For 70% of the patients, anti-epileptic medication is a sufficient treatment. The other 30% have the option to have epilepsy surgery where a region, the epileptogenic zone in the brain causing the seizures, is removed. Before the surgery, a hypothesis on this location can be made with magnetoencephalography (MEG). MEG is a technique that non-invasively measures brain activity based on the field produced by neuronal currents in the brain. The origin of these signals from the brain can be reconstructed through source reconstruction requiring a forward and inverse model. The forward model requires a head model, this can be a spherical or a realistic model. A realistic head model is made with meshes and describes the geometry of the head more accurately than a sphere. Simulation studies show a more accurate reconstruction of brain activity with realistic head models. In this study, it is evaluated whether this is also true in a clinical setting. The effect is expected mostly in the temporal lobe because its geometry deviates from a sphere. The following research question is answered: "Does a more detailed head model help to better detect epileptiform discharges from MEG signals in the temporal lobe in a clinical setting?" Since it is found in literature that the results will differ mostly in the temporal lobe, patients with hippocampal epilepsy are included. With stereo electroencephalography data, patients are selected and this resulted in seven patients. The Signal to Noise Ratio (SNR) (quantified as z-scores) for all peaks obtained with both the spherical and realistic head models were computed. For the realistic model, meshes of the head needed to be constructed and used to reconstruct the signals to the sources. These meshes were canonical meshes warped to the individual MRI. Peaks were manually identified, and the maximum (over channels and time) z-scores were calculated around these timestamps. A Wilcoxon signed-rank test was used for statistical differences between the SNR of the peaks that are reconstructed with both models. The results show no clear significant differences. On average, the spherical head model had a slightly higher SNR compared to the realistic head model (4.06 ± 0.86 vs 3.98 ± 0.87), but the difference was only significant for one out of seven patients. The steps to run the realistic head model are implemented in a MATLAB script and can be used for other patients. This study showed the effect of using a realistic head model instead of a spherical model on the SNR of clinical MEG data. No significant differences were found, showing that the realistic head model does not necessarily improve the source reconstruction. Only the z-scores, or SNR, were used to make the comparison, and it might add information to also look into the localization of the sources. Limitations of the study are the absence of a clear gold standard for localization, default conductivities, and a canonical mesh. Future studies could also quantify the effect of different mesh sizes or the number of layers.
Item Type:Essay (Master)
Amsterdam UMC, location VUmc, Amsterdam, Netherlands
Faculty:TNW: Science and Technology
Programme:Biomedical Engineering MSc (66226)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page