Synthetic Medical Images for Robust, Privacy-Preserving Training of Artificial Intelligence: Application to Retinopathy of Prematurity Diagnosis

This study uses progressively growing GANs (PGANs) to generate synthetic retinal vessel maps (RVMs) for training convolutional neural networks (CNNs) to detect plus disease in retinopathy of prematurity. CNNs trained on synthetic RVMs performed as well as or better than those trained on real RVMs, potentially due to the increased diversity of synthetic data. The paper highlights the privacy-preserving benefits of synthetic data, as it reduces the risk of sharing sensitive medical images, and suggests synthetic datasets can enhance small real datasets. To read the complete article click on the link... https://pubmed.ncbi.nlm.nih.gov/36249693/