Abstract
STS are rare malignant disorders, and numerous histological subtypes exist that impede the diagnostic accuracy due to small datasets that are annotated by experts at institutions. The author presents a priva cy-conserving federated learning architecture in this paper that enables the collaborative. training on a model across multiple medical institutions without breaching patient privacy. Our is an algorithm, which integrates the privacy of differentials, the homomorphic encryption, and the secure multi-party. compu tation. The experiments of multi-institutional histopathology data of large scale confirm. that federated learning can equally achieve diagnostics, as centralized models. It eliminates the privacy threat and, (87.3% vs. 89.1% accuracy) besides, it is more precise. The framework addresses non-IID. data distri bution, communication efficiency, and resistance to attack vectors and demonstrates federal federated. learning to be a feasible solution to the diagnostics of rare cancer and satisfies the regulations at the same time.and governance enhancement.
To Read or Download the Article PDF