Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Think Twice Before Selection: Federated Evidential Active Learning for Medical Image Analysis with Domain Shifts

Jiayi Chen, Benteng Ma, Hengfei Cui, Yong Xia

TL;DR

This work tackles label efficiency in federated medical imaging under realistic domain shifts by introducing Federated Evidential Active Learning (FEAL). FEAL represents predictions as Dirichlet distributions, enabling explicit modeling of both aleatoric and epistemic uncertainties, and uses calibrated sampling to combine global and local model insights. The method incorporates a diversity-relaxation strategy and evidence-based training to improve data valuation and model learning, achieving state-of-the-art results across five real multi-center datasets for both classification and segmentation. Extensive ablations confirm the value of uncertainty calibration, diversity control, and evidential training, while OCTA-500 experiments demonstrate robust performance with limited annotations and strong practical impact for privacy-preserving, collaborative healthcare research.

Abstract

Federated learning facilitates the collaborative learning of a global model across multiple distributed medical institutions without centralizing data. Nevertheless, the expensive cost of annotation on local clients remains an obstacle to effectively utilizing local data. To mitigate this issue, federated active learning methods suggest leveraging local and global model predictions to select a relatively small amount of informative local data for annotation. However, existing methods mainly focus on all local data sampled from the same domain, making them unreliable in realistic medical scenarios with domain shifts among different clients. In this paper, we make the first attempt to assess the informativeness of local data derived from diverse domains and propose a novel methodology termed Federated Evidential Active Learning (FEAL) to calibrate the data evaluation under domain shift. Specifically, we introduce a Dirichlet prior distribution in both local and global models to treat the prediction as a distribution over the probability simplex and capture both aleatoric and epistemic uncertainties by using the Dirichlet-based evidential model. Then we employ the epistemic uncertainty to calibrate the aleatoric uncertainty. Afterward, we design a diversity relaxation strategy to reduce data redundancy and maintain data diversity. Extensive experiments and analysis on five real multi-center medical image datasets demonstrate the superiority of FEAL over the state-of-the-art active learning methods in federated scenarios with domain shifts. The code will be available at https://github.com/JiayiChen815/FEAL.

Think Twice Before Selection: Federated Evidential Active Learning for Medical Image Analysis with Domain Shifts

TL;DR

This work tackles label efficiency in federated medical imaging under realistic domain shifts by introducing Federated Evidential Active Learning (FEAL). FEAL represents predictions as Dirichlet distributions, enabling explicit modeling of both aleatoric and epistemic uncertainties, and uses calibrated sampling to combine global and local model insights. The method incorporates a diversity-relaxation strategy and evidence-based training to improve data valuation and model learning, achieving state-of-the-art results across five real multi-center datasets for both classification and segmentation. Extensive ablations confirm the value of uncertainty calibration, diversity control, and evidential training, while OCTA-500 experiments demonstrate robust performance with limited annotations and strong practical impact for privacy-preserving, collaborative healthcare research.

Abstract

Federated learning facilitates the collaborative learning of a global model across multiple distributed medical institutions without centralizing data. Nevertheless, the expensive cost of annotation on local clients remains an obstacle to effectively utilizing local data. To mitigate this issue, federated active learning methods suggest leveraging local and global model predictions to select a relatively small amount of informative local data for annotation. However, existing methods mainly focus on all local data sampled from the same domain, making them unreliable in realistic medical scenarios with domain shifts among different clients. In this paper, we make the first attempt to assess the informativeness of local data derived from diverse domains and propose a novel methodology termed Federated Evidential Active Learning (FEAL) to calibrate the data evaluation under domain shift. Specifically, we introduce a Dirichlet prior distribution in both local and global models to treat the prediction as a distribution over the probability simplex and capture both aleatoric and epistemic uncertainties by using the Dirichlet-based evidential model. Then we employ the epistemic uncertainty to calibrate the aleatoric uncertainty. Afterward, we design a diversity relaxation strategy to reduce data redundancy and maintain data diversity. Extensive experiments and analysis on five real multi-center medical image datasets demonstrate the superiority of FEAL over the state-of-the-art active learning methods in federated scenarios with domain shifts. The code will be available at https://github.com/JiayiChen815/FEAL.
Paper Structure (50 sections, 19 equations, 21 figures, 11 tables, 4 algorithms)

This paper contains 50 sections, 19 equations, 21 figures, 11 tables, 4 algorithms.

Table of Contents

  1. Introduction
  2. Related Work
  3. Federated Learning with Domain Shifts
  4. AL Methods
  5. FAL Methods
  6. Methodology
  7. Problem Formulation
  8. Dirichlet-based Evidential Model in FAL
  9. Calibrated Evidential Sampling
  10. Evidential Model Learning
  11. Experiments
  12. Experimental Settings
  13. Results
  14. Discussion
  15. Conclusion and Social Impact
  16. ...and 35 more sections

Figures (21)

  • Figure 1: FAL scheme
  • Figure 2: KDE of energy score
  • Figure 3: $p$-value
  • Figure 5: Illustration of the proposed FEAL method. (a) Overview of FEAL. (b) Illustration of CES module, including uncertainty calibration and diversity relaxation.
  • Figure 6: Comparison of FAL methods in medical image classification. (a)-(c) and (d)-(f) depict the results of the Fed-ISIC and Fed-Camelyon datasets, respectively. Performance enhancements over the second-best method in each FAL round are emphasized in red text.
  • ...and 16 more figures