Effect of MRI Defacing on EEG Forward and Inverse Modeling

Recent years have seen improvements in facial recognition software, which has increased the risk of re-identification of patients' structural neuroimaging scans, e.g. obtained from magnetic resonance imaging (MRI). Thus, an anonymization procedure known as defacing has become the norm when publicly sharing patients' scans. Defacing removes some of the facial features from the data, making it improbable to re-identify a patient from a 3D rendering of the image. However, certain tasks, such as localization of the sources of electroencephalographic (EEG) signals, require the creation of individual electrical volume conductor models of the human head from structural MRI data. Defacing could affect the co-registeration of MRI and EEG sensor positions and, more importantly, the model of electrical current flow itself. This study quantifies and maps the effect of defacing on individual volume conductor models and the localization accuracy of inverse solutions based on these models in a sample of ten subjects with known ground-truth (non-defaced) anatomy. Boundary and finite element modeling approaches (B/FEM) are compared.Clinical relevance- This study enables clinicians, e.g. epileptologists, and neuroscientists to gauge the EEG source localization error that is incurred due to using defaced MRI data for forward model construction, and identifies the regions most prone to mislocalization. Based on the results of the study, it is recommended to construct EEG volume conductor models before defacing if accurate source localization is desired.