Enable the Right to be Forgotten with Federated Client Unlearning in Medical Imaging

TL;DR

This paper introduces a novel federated client unlearning framework for medical imaging that enables clients to erase their data influence efficiently, maintaining model performance and privacy, with significant speed improvements over retraining.

Contribution

The paper presents the first federated unlearning framework tailored for medical imaging, incorporating feature-level unlearning and frequency-guided memory preservation.

Findings

Outperforms existing FU frameworks on medical datasets.

Achieves 10-15 times faster unlearning than retraining from scratch.

Maintains model accuracy and privacy during unlearning process.

Abstract

The right to be forgotten, as stated in most data regulations, poses an underexplored challenge in federated learning (FL), leading to the development of federated unlearning (FU). However, current FU approaches often face trade-offs between efficiency, model performance, forgetting efficacy, and privacy preservation. In this paper, we delve into the paradigm of Federated Client Unlearning (FCU) to guarantee a client the right to erase the contribution or the influence, introducing the first FU framework in medical imaging. In the unlearning process of a client, the proposed model-contrastive unlearning marks a pioneering step towards feature-level unlearning, and frequency-guided memory preservation ensures smooth forgetting of local knowledge while maintaining the generalizability of the trained global model, thus avoiding performance compromises and guaranteeing rapid post-training. We evaluated our FCU framework on two public medical image datasets, including Intracranial hemorrhage diagnosis and skin lesion diagnosis, demonstrating that our framework outperformed other state-of-the-art FU frameworks, with an expected speed-up of 10-15 times compared with retraining from scratch. The code and the organized datasets can be found at: https://github.com/dzp2095/FCU.