Fair Federated Medical Image Segmentation via Client Contribution Estimation

Abstract

How to ensure fairness is an important topic in federated learning (FL). Recent studies have investigated how to reward clients based on their contribution (collaboration fairness), and how to achieve uniformity of performance across clients (performance fairness). Despite achieving progress on either one, we argue that it is critical to consider them together, in order to engage and motivate more diverse clients joining FL to derive a high-quality global model. In this work, we propose a novel method to opti- mize both types of fairness simultaneously. Specifically, we propose to estimate client contribution in gradient and data space. In gradient space, we monitor the gradient direction differences of each client with respect to others. And in data space, we measure the prediction error on client data us- ing an auxiliary model. Based on this contribution estima- tion, we propose a FL method, federated training via contri- bution estimation (FedCE), i.e., using estimation as global model aggregation weights. We have theoretically analyzed our method and empirically evaluated it on two real-world medical datasets. The effectiveness of our approach has been validated with significant performance improvements, better collaboration fairness, better performance fairness, and comprehensive analytical studies.

Method

FedCE Framework

Figure 1: The proposed FedCE framework with client contribution estimation mechanism.

Client Contribution Estimation

Algorithm

Results

Sample distribution

Figure 2. Pixel intensity distributions. Left denotes samples from retinal fundus dataset and right ones are from prostate MRI.

Table 1: Performance comparison using Dice score on image segmentation datasets of retinal fundus images and prostate MRI.

Collaboration fairness

Table 2: Client contribution estimation comparison by comparing the results of leave-one-out with that of our method and others.

Performance fairness

Table 3: Fairness comparison with our method and others.

Distributional robustness

Figure 4. Study to validate the distribution shift robustness of our contribution estimation metric. The y-axis denotes the relative weight percentage, and \Delta denotes the differences.

Ablation study

Figure 5. Ablation study on effects of two separate contribution quantification metrics and their combination on two datasets.

Free rider detection

Figure 6: Free rider study. Each row denotes an independent fed- erated training procedure, and the y-axis indicates which client is the free rider. A free rider is detected with a high value.

BibTeX

@inproceedings{jiang2023fedce,
  title={Fair Federated Medical Image Segmentation via Client Contribution Estimation},
  author={Jiang, Meirui and Roth, Holger R and Li, Wenqi and Yang, Dong and Zhao, Can and Nath, Vishwesh and Xu, Daguang and Dou, Qi and Xu, Ziyue},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  year={2023}
}