Privacy-Preserving Healthcare Diagnostics: A Comparative Study of Horizontal and Vertical Federated Learning Approaches

Abstract

Federated learning (FL) is a novel method of healthcare diagnostics that allows training cross -institutional collaborative models without violating data privacy. This paper will compare and contrast the two approaches of vertical and horizontal federated learning to detect diseases such as pneumonia based on the use of chest X -rays. Horizontal FL applies to the same data of multiple institutions, and vertical FL applies to complementary feature sets of overlapping groups of patients. As experimental findings indicate, horizontal FL is more efficient than vertical in a homogeneous data distribution, but vertical FL i s more effective in cases of heterogeneity in features. The results show that FL has the ability to enable scalable and privacy-sensitive diagnostic applications in diverse healthcare environments.