Medical Dataset Preparation and Privacy Preservation for Improving the Healthcare Facilities Using Federated Learning Approach

Abstract

As a result of recent advances in deep learning, several breakthrough stories in modern medical diagnostics with data-driven insights into improving healthcare facilities' quality of treatment have arisen. Deep learning methods that perform well are significantly data-driven. As more data is trained, the deep learning model's performance becomes much more robust and generalizable. On the other hand, collecting medical data in a central storage system to train effective deep learning models raises concerns about privacy, ownership, and regulatory compliance. Federated learning overcomes the previous difficulties by using a deep learning model which is shared and a centralized aggregating platform. On the other side, patient data resides with the local party, assuring data confidentiality and data security. First, we give a comprehensive, up-to-date survey of federated learning research in healthcare applications. Next, we propose a solution for preparation of the medical dataset for federated learning approach from publicly available medical repositories and then apply federated averaging(FedAvg) and FedProx algorithm for aggregating across clients without accessing local private data.

Ähnliche Arbeiten