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

2026-05-11

AI Summary

This work addresses the performance degradation of medical image segmentation models in federated learning settings caused by heterogeneous and unreliable client-side annotations. To mitigate this issue, the authors propose a novel framework that integrates split federated learning with collaborative learning. The approach leverages a global teacher model to guide local student models in identifying and correcting inaccurate labels, enhanced by consistency regularization, a learnable weighting module, a difficulty-guided annotation perturbation strategy, and an adaptive loss balancing mechanism. These components collectively improve the model’s robustness to low-quality and heterogeneous annotations. Extensive experiments on multiple medical image segmentation datasets—containing both synthetic noise and real-world labeling errors—demonstrate that the proposed method significantly outperforms seven state-of-the-art approaches, achieving superior segmentation accuracy and stability.

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.

Problem

• medical image segmentation
• inaccurate labels
• label heterogeneity
• federated learning
• annotation errors

Innovation

• SplitFed-CL
• co-learning
• annotation refinement
• consistency regularization
• difficulty-guided noise