Responsible AI Seminar: Sharing synthetic medical images - a way to circumvent GDPR?

Hybrid talk with Anders Eklund (Linköping University): Deep learning research in medical imaging is currently impeded by the lack of available training data. There is not a lack of data per se, but the major part of medical images stored in hospital databases cannot be used for research, due to perceived and actual regulatory barriers. Recent research has shown that generative adversarial networks (GANs) can be trained to synthesize very realistic images (e.g. thispersondoesnotexist.com).

In my presentation, I will focus on how GANs can be used for synthesizing medical images, 2D and 3D, and how this relates to data sharing. Can synthetic medical images be seen as anonymized, as they do not belong to a specific person, and therefore be shared freely? Can AI models be successfully trained using synthetic images?

This seminar can also be attended physically at the Neurobiology Research Unit (entrance 8, floor 5, conference room), 6-8 Inge Lehmanns Vej, Rigshospitalet, building 8057, DK-2100 Copenhagen, Denmark.

Initiators:Aasa FeragenMelanie Ganz and Sune Hannibal Holm from the DFF-funded project Bias and Fairness in Medicine.

Tidspunkt

fre 17 sep 21
11:00 - 12:00

Arrangør

DTU Compute

Hvor

Hybrid: Zoom-link

This seminar can also be attended physically at the Neurobiology Research Unit (entrance 8, floor 5, conference room), 6-8 Inge Lehmanns Vej, Rigshospitalet, building 8057, DK-2100 Copenhagen, Denmark