PhD ‘Deepfakes’ in radiology? Synthetic imaging data medical AI regulation

Artificial intelligence is increasingly used in medical imaging, but the evidence required to demonstrate that these systems are safe and reliable is still evolving. Synthetic imaging data may help address limited access to representative clinical datasets, for example by augmenting training data or supporting model evaluation. However, it is currently unclear under which conditions synthetic data can be considered sufficiently representative, unbiased, privacy-preserving, and scientifically valid for (premarket) regulatory purposes.

In this PhD project, you will investigate when and how synthetic medical imaging data can be used as evidence in the development and regulatory evaluation of medical device AI. The project is part of COMPASS4MD, a Horizon Europe consortium developing new methodologies for evaluating clinical evidence for medical devices.

Your work will combine technical evaluation of synthetic imaging data with regulatory science and research methodology. You will study how synthetic imaging data are currently used in research and regulatory submissions, experimentally assess their quality and fitness for different purposes, and translate these findings into evidence requirements for regulatory use.

You will work on questions such as:

• How representative are synthetic images compared with real clinical data?
• Which metrics are meaningful for assessing fidelity and clinical validity?
• How do privacy protection, bias and representativeness interact?
• When can synthetic data appropriately be used for training augmentation, and when might they be suitable for external validation?
• How should evidence requirements differ between detection, segmentation, classification and quantification tasks?
• What additional limitations arise for rare and ultra-rare diseases?

The technical work will include analysis of medical imaging datasets and evaluation of synthetic image generation methods. Depending on the specific research questions, this may include image quality and distributional analyses, evaluation of downstream AI performance, assessment of bias and representativeness, and privacy-related analyses. You will work closely with researchers at the University of Crete, who lead the technical characterisation of synthetic imaging within COMPASS4MD.

Alongside this technical work, you will conduct a structured review of the scientific and regulatory landscape, interview stakeholders including regulators, notified bodies and AI developers, and contribute to a multi-round Delphi study. The final aim is to develop minimum evidence standards for the use of synthetic imaging data in regulatory submissions for diagnostic medical AI. These standards will distinguish between different functions of synthetic data, AI task types and disease prevalence.

You will be based at Radboudumc and jointly supervised by Dr Merel Huisman and Dr Michail Klontzas at the University of Crete. The project therefore offers the opportunity to work at the intersection of medical imaging AI, methodology and European regulatory science within an international consortium.

You will be based in the Department of Medical Imaging at Radboud University Medical Center (Radboudumc), embedded in the Diagnostic Image Analysis Group (DIAG). DIAG is one of Europe’s leading research groups in medical image analysis, with researchers across Radiology and Nuclear Medicine, Pathology, and Cardiology. The group has a strong track record of publishing in high-impact journals. The working atmosphere is collaborative, international, and focused on clinical impact.

This project connects technical research on synthetic medical imaging data with clinical evidence and regulatory science. Within COMPASS4MD, a Horizon Europe consortium developing new methodologies for evaluating clinical evidence for medical devices, you will investigate when synthetic imaging data can support the development and regulatory evaluation of medical AI. You will combine technical evaluation with research on the evidence requirements for responsible regulatory use.

You will be jointly supervised by Dr M. Huisman (PI), at Radboudumc and Dr Michail Klontzas at the University of Crete. Dr Huisman is an attending cardiovascular and musculoskeletal radiologist and Vice President Elect of EuSoMII, with expertise in clinical AI validation, post-market surveillance, and standardisation of AI as a regulated clinical technology. Dr Klontzas and his team lead the technical characterisation of…