Distorting anatomy to test MEG models and metrics

Abstract

Current flow that gives rise to non-invasive Magnetoencephalographic (MEG) data derives predominantly from pyramidal neurons oriented orthogonal to the cortical surface. The estimate of current flow based on extra-cranial magnetic fields is a well-known ill-posed problem; however, this current distribution must depend on anatomy. In other words, a veridical estimate of current flow should discriminate between true and distorted versions of the brain. Here, we make use of advances in diffeomorphic brain shape modelling to construct a set of parametrically deformable cortical surfaces. We use a latent space of 100 components to construct cortical surfaces that are representative of the population. We show how these geometric distortions can be used to quantify the performance of MEG source reconstruction algorithms and metrics of fit.