Researchers at MedUni Wien have developed a generative AI system that can create synthetic medical images, improving the accuracy of AI-powered diagnostic systems. This breakthrough, published in the European Journal of Nuclear Medicine and Molecular Imaging, addresses the limitations of traditional clinical datasets, which are often small, restricted by privacy regulations, or lack representation of certain subgroups.
The study involved training a generative AI model on over 9,000 scans from routine clinical examinations. The model then generated a synthetic dataset mirroring the characteristics of real medical images, but without containing any patient-specific information. This allows for ethical use of the data for research and development of AI-based diagnostic tools.
Independent evaluations confirmed the quality and relevance of the synthetic data. When used to train an AI system for detecting cardiac amyloidosis or bone metastases, the system demonstrated significant improvements in diagnostic accuracy, proving the effectiveness of this approach.
Generative AI creates synthetic medical images, addressing limitations of traditional datasets.
The synthetic data allows for ethical research and development of AI-based diagnostic tools.
The study demonstrates improved diagnostic accuracy in AI systems trained with synthetic data.