Decentralized Federated Learning for Privacy-Preserving Cardiac Image Analysis

Heart-related problems are the number one killer in today’s world, so it is necessary to have early screening methods with imaging systems like X-ray, MRI, and CT. Medical data is very confidential and is subject to privacy laws that limit its use outside hospitals. Therefore, it would be difficult to build efficient machine learning algorithms based on centralized data sources. Federated Learning technology allows for a collaborative effort between different parties to perform machine learning-based analyses in the field of cardiology in a confidential manner. A scenario involving multiple clients is constructed by splitting the data into several parts, with each part treated as a separate client. The clients train independently and upload their results to a central server. The federated learning paradigm works successfully in this scenario, and data confidentiality was maintained. Real implementation is not carried out; however, the theoretical idea was proven efficiently.