Federated Learning for Securing Glomeruli Detection in Digital Pathology

Abstract

This chapter addresses the imperative for secure and precise glomeruli detection in histopathological images, crucial for diagnosing renal conditions. Using the YOLOv3 object detection algorithm, it integrates federated learning to ensure data security. A custom dataset, annotated with XML labels from histopathological images of sclerosed and normal glomeruli, is created. Initially, a traditional YOLOv3 model achieved 98.55% accuracy but posed privacy risks with centralized data storage. Federated learning decentralizes data across clients, preserving privacy and achieving 98.79% accuracy. Employing cryptographic techniques for data transmission security, this chapter demonstrates federated learning's robustness in medical image analysis. A comparative analysis with traditional methods highlights federated learning's advantages in data security and collaborative learning, showcasing its transformative potential in digital pathology.