Federated Medical Image Segmentation under Real-World Label Noise: A Benchmark Suite for Noisy Label Learning Method Selection

The article introduces a benchmark suite for Federated Noisy Label Learning (FNLL) aimed at improving medical image segmentation in the presence of real-world label noise. It includes diverse datasets and client-noise scenarios, enabling systematic assessment of FNLL methods in realistic environments. This suite is significant for practitioners as it provides a structured approach to evaluate and select methods for handling label imperfections in federated learning settings, thereby enhancing the robustness of medical image segmentation applications.