SplitFed-CL: A Split Federated Co-Learning Framework for Medical Image Segmentation with Inaccurate Labels

TL;DR

SplitFed-CL is a novel federated co-learning framework that improves medical image segmentation accuracy and robustness by handling unreliable labels through a teacher-student model, consistency regularization, and adaptive loss balancing.

Contribution

The paper introduces SplitFed-CL, a new framework combining federated and split learning with a teacher-student model for label refinement in medical segmentation.

Findings

Outperforms seven state-of-the-art baselines in segmentation quality.

Demonstrates robustness to synthetic and real-world annotation errors.

Improves segmentation accuracy with unreliable labels.

Abstract

Split Federated Learning (SplitFed) combines federated and split learning to preserve privacy while reducing client-side computation. However, in medical image segmentation, heterogeneous label quality across clients can significantly degrade performance. We propose SplitFed-CL, a co-learning framework where a global teacher guides local students to detect and refine unreliable annotations. Reliable labels supervise training directly, while unreliable labels are corrected via weighted student--teacher refinement. SplitFed-CL further incorporates consistency regularization for robustness to input perturbations and a trainable weighting module to balance loss terms adaptively. We also introduce a novel difficulty guided strategy to simulate human like boundary centric annotation errors, where the degree of perturbation is governed by shape complexity and the associated annotation difficulty. Experiments on two multiclass segmentation datasets with controlled synthetic noise, together with a binary segmentation dataset containing real-world annotation errors, demonstrate that SplitFed-CL consistently outperforms seven state-of-the-art baselines, yielding improved segmentation quality and robustness.