Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

FedIA: Federated Medical Image Segmentation with Heterogeneous Annotation Completeness

Yangyang Xiang, Nannan Wu, Li Yu, Xin Yang, Kwang-Ting Cheng, Zengqiang Yan

TL;DR

This work addresses the challenge of heterogeneous annotation completeness in federated medical image segmentation by treating incomplete labels as noisy data. It introduces FedIA, a four-stage framework that first learns a noise-robust warm-up model, then estimates per-client annotation completeness $\hat{a}_k$, applies completeness-aware aggregation via weights $w_k^t$, and finally performs IoU-based, confidence-filtered annotation corrections to refine data quality. Key contributions include a formal problem formulation for annotation-heterogeneous FL, a practical completeness estimation and weighted aggregation strategy, and a client-wise correction mechanism driven by IoU trends, all validated on MS brain MRI lesion and COVID-19 lung lesion datasets with superior performance over state-of-the-art baselines. The approach offers a practical pathway to robust federated segmentation in real-world clinical settings where labeling effort varies across institutions.

Abstract

Federated learning has emerged as a compelling paradigm for medical image segmentation, particularly in light of increasing privacy concerns. However, most of the existing research relies on relatively stringent assumptions regarding the uniformity and completeness of annotations across clients. Contrary to this, this paper highlights a prevalent challenge in medical practice: incomplete annotations. Such annotations can introduce incorrectly labeled pixels, potentially undermining the performance of neural networks in supervised learning. To tackle this issue, we introduce a novel solution, named FedIA. Our insight is to conceptualize incomplete annotations as noisy data (i.e., low-quality data), with a focus on mitigating their adverse effects. We begin by evaluating the completeness of annotations at the client level using a designed indicator. Subsequently, we enhance the influence of clients with more comprehensive annotations and implement corrections for incomplete ones, thereby ensuring that models are trained on accurate data. Our method's effectiveness is validated through its superior performance on two extensively used medical image segmentation datasets, outperforming existing solutions. The code is available at https://github.com/HUSTxyy/FedIA.

FedIA: Federated Medical Image Segmentation with Heterogeneous Annotation Completeness

TL;DR

This work addresses the challenge of heterogeneous annotation completeness in federated medical image segmentation by treating incomplete labels as noisy data. It introduces FedIA, a four-stage framework that first learns a noise-robust warm-up model, then estimates per-client annotation completeness , applies completeness-aware aggregation via weights , and finally performs IoU-based, confidence-filtered annotation corrections to refine data quality. Key contributions include a formal problem formulation for annotation-heterogeneous FL, a practical completeness estimation and weighted aggregation strategy, and a client-wise correction mechanism driven by IoU trends, all validated on MS brain MRI lesion and COVID-19 lung lesion datasets with superior performance over state-of-the-art baselines. The approach offers a practical pathway to robust federated segmentation in real-world clinical settings where labeling effort varies across institutions.

Abstract

Federated learning has emerged as a compelling paradigm for medical image segmentation, particularly in light of increasing privacy concerns. However, most of the existing research relies on relatively stringent assumptions regarding the uniformity and completeness of annotations across clients. Contrary to this, this paper highlights a prevalent challenge in medical practice: incomplete annotations. Such annotations can introduce incorrectly labeled pixels, potentially undermining the performance of neural networks in supervised learning. To tackle this issue, we introduce a novel solution, named FedIA. Our insight is to conceptualize incomplete annotations as noisy data (i.e., low-quality data), with a focus on mitigating their adverse effects. We begin by evaluating the completeness of annotations at the client level using a designed indicator. Subsequently, we enhance the influence of clients with more comprehensive annotations and implement corrections for incomplete ones, thereby ensuring that models are trained on accurate data. Our method's effectiveness is validated through its superior performance on two extensively used medical image segmentation datasets, outperforming existing solutions. The code is available at https://github.com/HUSTxyy/FedIA.
Paper Structure (16 sections, 7 equations, 3 figures, 3 tables)

This paper contains 16 sections, 7 equations, 3 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Preliminaries and Overview
  4. Annotation Completeness Estimation
  5. Annotation Completeness-Aware Aggregation
  6. Client-wise Adaptive Correction
  7. Experiments
  8. Datasets and Implementation Details
  9. Datasets.
  10. Incomplete Annotation Generation
  11. Implementation Details.
  12. Comparison with State-of-the-art Methods
  13. Ablation Study
  14. Component-wise Study.
  15. Impact of the Early Learning Round $T$.
  16. ...and 1 more sections

Figures (3)

  • Figure 1: Heterogeneity in annotation completeness among clients. Red and blue solid lines represent the boundaries of marked lesions and unmarked lesions, respectively.
  • Figure 2: Overview of the proposed FedIA. The first stage is the early learning phase, the global model is updated by FedAvg FedAvg. The second is the modification stage, re-weighting each client by calculating its annotation completeness rate and correcting incomplete annotations synchronously. In the last stage, local models are trained with the corrected labels and aggregated for federated updating through FedAvg FedAvg.
  • Figure 3: Qualitative comparison on MS where others represents FedAvg, ELR, ADELE, and FedCorr failing to segment any lesion. Red, blue and green color show the prediction of true-positive, false-negative and false-positive regions, respectively.