Privacy-stealing Approach in Distributed IoMT Systems

Model or parameter sharing has been commonly used in many distributed learning architectures, e.g., federated learning. However, such a sharing mechanism may result in the risk of privacy leakage to the third party. In this paper, we represent an approach to medical image privacy stealing in the distributed Internet of Medical Things (IoMT) systems. Experiment results based on the real-world dataset show that, based on our approach, key characteristics of original medical images could be recovered in a third party, i.e., aggregation server.