Refacing: reconstructing anonymized facial features using GANs
This addresses privacy risks in medical data sharing for patients and researchers, highlighting vulnerabilities in current anonymization methods.
The study tackled the problem of medical image anonymization by using CycleGAN to reconstruct facial features from anonymized T1 MR images, finding that face blurring is inadequate and face removal is partially reversible.
Anonymization of medical images is necessary for protecting the identity of the test subjects, and is therefore an essential step in data sharing. However, recent developments in deep learning may raise the bar on the amount of distortion that needs to be applied to guarantee anonymity. To test such possibilities, we have applied the novel CycleGAN unsupervised image-to-image translation framework on sagittal slices of T1 MR images, in order to reconstruct facial features from anonymized data. We applied the CycleGAN framework on both face-blurred and face-removed images. Our results show that face blurring may not provide adequate protection against malicious attempts at identifying the subjects, while face removal provides more robust anonymization, but is still partially reversible.