AIF-SFDA: Autonomous Information Filter-driven Source-Free Domain Adaptation for Medical Image Segmentation

TL;DR

This work addresses domain shifts in medical image segmentation under strict privacy that prevents access to source data. It introduces AIF-SFDA, a two-stage, target-data-only SFDA method that learns a frequency-based information filter guided by Information Bottleneck and Self-supervision to autonomously decouple DVI from DII, using losses such as $ ext{PL}$, $ ext{MI}$, and feature consistency. Key contributions include the autonomous learnable filter, IB-constrained DVI reduction, SS-based DII preservation, and extensive cross-domain validation on retinal vessel and ultrasound cartilage segmentation, where AIF-SFDA achieves superior generalization to unseen targets. This approach enables robust, privacy-preserving medical image segmentation across modalities, providing practical impact for real-world clinical deployments.

Abstract

Decoupling domain-variant information (DVI) from domain-invariant information (DII) serves as a prominent strategy for mitigating domain shifts in the practical implementation of deep learning algorithms. However, in medical settings, concerns surrounding data collection and privacy often restrict access to both training and test data, hindering the empirical decoupling of information by existing methods. To tackle this issue, we propose an Autonomous Information Filter-driven Source-free Domain Adaptation (AIF-SFDA) algorithm, which leverages a frequency-based learnable information filter to autonomously decouple DVI and DII. Information Bottleneck (IB) and Self-supervision (SS) are incorporated to optimize the learnable frequency filter. The IB governs the information flow within the filter to diminish redundant DVI, while SS preserves DII in alignment with the specific task and image modality. Thus, the autonomous information filter can overcome domain shifts relying solely on target data. A series of experiments covering various medical image modalities and segmentation tasks were conducted to demonstrate the benefits of AIF-SFDA through comparisons with leading algorithms and ablation studies. The code is available at https://github.com/JingHuaMan/AIF-SFDA.

Figures (6)

• Figure 1: To develop an autonomous information filter in SFDA scenarios, we utilize IB-constrained mutual information constraint to reduce DVI in the image information while preserving DII through SS-constrained guidance.
• Figure 2: The architecture of our proposed AIF-SFDA.
• Figure 3: Qualitative results for retinal vessel segmentation, where true positive pixels are colored in magenta, false positive pixels in red, and false negative pixels in blue.
• Figure 4: Qualitative results for cartilage segmentation.
• Figure 5: Information filter processing for different tasks. (a) and (b) show the filtering process for vessel and optic disc (OD) segmentation, respectively. (I. original image, II. filtered image, III. filter map, IV. ground truth)