Source independent multiple-domain adaptation for knee osteoarthritis cartilage and meniscus segmentation in clinical magnetic resonance imaging

Background: Generalized knee tissue segmentation, such as cartilage and meniscus in magnetic resonance imaging (MRI), plays a vital role in the clinical assessment of knee osteoarthritis (OA). However, domain variability between MRI datasets poses a significant challenge for robust segmentation methods in real-world clinical applications. Existing unsupervised domain adaptation (UDA) approaches, which rely on one-to-one assumptions between source and target domains, often fail to preserve knee tissues such as cartilage and meniscus, which is critical for OA diagnosis in diverse clinical settings. Method: We propose a source-independent segmentation approach tailored for multi-domain knee MRI datasets. Our method emphasizes knee tissue regions to reduce domain gaps and label inconsistencies. By introducing a stepwise adaptation strategy, segmentation performance is refined progressively from intermediate domains to the final target domain. Pseudo-label attention mechanisms are integrated into the adaptation pipeline, enabling iterative fine-tuning of domain-specific segmentations while leveraging unidirectional generative adversarial networks (GANs) to enhance tissue-specific adaptation. This iterative training process ensures the generation of reliable pseudo-labels, improving segmentation accuracy in diverse clinical MRI datasets. Results: We demonstrate the effectiveness of our approach on the OA initiative dataset as the source domain and self-collected, T1-weighted fast field echo as the intermediate domain and three-dimensional fast spin echo as the final target domain. Our method achieves an average dice scores of 0.8701 and 0.7990 for source and target domain respectively, surpassing typical UDA methods explored in our experiments. Conclusion: The experiments conducted on clinical MRI data, spanning OA severity from healthy knees to KL Grades 1–4, validate the effectiveness of the proposed domain adaptation method in precise segmentation of cartilage and meniscus.