Efficient and Privacy-Enhanced Federated Learning for Medical Imaging in Resource-Limited Environments

Abstract

Deep learning's quick development has transformed computer-aided laboratory services and healthcare by facilitating data-driven decision-making for better patient care. However, privacy, data ownership, and regulatory compliance pose serious problems for the centralized aggregation of medical data. A potential approach that enables several healthcare organizations to work together to train machine learning models without exchanging raw patient data is federated learning (FL). The effectiveness of FL in resolving issues including data silos, class imbalance, and non-IID data distributions is examined in this paper's systematic assessment of FL applications in healthcare. In order to uncover systemic issues that affect FL adoption in actual healthcare settings and to highlight significant methodological breakthroughs, we examined 89 research publications published between January 2015 and February 2023. Our results highlight the necessity for stronger aggregation approaches, effective communication tactics, and better privacy-preserving methodologies to improve FL performance in clinical data processing and medical imaging. The paper ends with suggestions to direct further investigation and the creation of FL frameworks that are optimal for use in healthcare applications.