Federated learning for analysis of medical images: a survey

Machine learning models trained in medical imaging can help in the early detection, diagnosis, and prognosis of the disease. However, it confronts two major obstacles: deep learning models require access to a substantial amount of imaging data, which is a hard constraint, and the patient data is private and sensitive, so it cannot be shared like 1 other imaging data in computer vision. Federated Learning (FL) offers an alternative by deploying many training models in a decentralized way. In recent years, various techniques that leverage FL for disease diagnosis have been introduced. Existing survey articles have analyzed and collated research about the use of FL in general. However, the particular component of medical imaging is ignored. The motivation behind this survey paper is to fill up the research gap by providing a comprehensive survey of FL techniques for medical imaging and various ways in which FL is employed to provide secure, accessible, and collaborative deep learning models for the medical imaging research community.