Fig. 1.

Outline of the simulation process. We use FreeSurfer to extract the pial and white matter surfaces for a single male subject. The pial and the white matter surfaces define the superficial and deep generative surface models. Activity is simulated by randomly selecting source locations from a list of 21401 possible patch centres. Data are generated using two different patch sizes; corresponding to either FWHM = 5 mm or FWHM = 10 mm (top left inset panel). Sources are simulated on both surfaces, followed by co-registration of each dataset to both cortical models. To add coregistration error, a random perturbation is added to the fiducial locations at this stage, either taking the form of rotation or pure translation of zero mean and specified standard deviation (0, 1, 2, 5, 10, 20 mm/degrees). Finally, we use the MSP algorithm to perform source reconstruction. This yields a free energy value as an outcome, which approximates the log model evidence, allowing us to compare between reconstructions based on the ‘correct’/‘incorrect’ surface models, as well as ‘correct’/‘incorrect’ patch sizes.