A growing amount of imaging data is made publicly available. While this is desirable for science and its reproducibility, privacy concerns increase. As the shape of a face can be recovered based on MR images, an increased number of studies remove the face from the data to prevent biometric identification. This defacing can, however, pose a challenge to existing post-processing pipelines e.g. brain volume assessment. This work investigates the impact of regenerating facial structures in defaced images on morphometry in a large cohort using a deep neural network. The results show that refacing can prevent volumetric errors induced by defacing.
This abstract and the presentation materials are available to members only; a login is required.