The paper "Learning Generalized Medical Image Segmentation from Decoupled Feature Queries" addresses the challenge of domain generalization in medical image segmentation, which requires models to learn from multiple source domains and perform well on unseen target domains. The authors propose a novel method called Decoupled Feature Query (DFQ) to address feature misalignment and improve generalization. DFQ leverages channel-wise decoupled deep features as queries, guided by cross-attention mechanisms, to learn a more expressive and generalized representation. A relaxed deep whitening transformation is introduced to decorrelate features, reducing channel redundancy and enhancing representation ability. The proposed method is integrated into Transformer segmentation models and demonstrates state-of-the-art performance on two benchmarks: optic cup/disk segmentation on fundus images and prostate segmentation on MRI. Extensive experiments show that the proposed method outperforms existing methods by at least 1.31% and 1.98% in Dice Similarity Coefficient (DSC) on the Fundus and Prostate benchmarks, respectively. The paper also includes ablation studies and visualized segmentation results to validate the effectiveness of the proposed approach.The paper "Learning Generalized Medical Image Segmentation from Decoupled Feature Queries" addresses the challenge of domain generalization in medical image segmentation, which requires models to learn from multiple source domains and perform well on unseen target domains. The authors propose a novel method called Decoupled Feature Query (DFQ) to address feature misalignment and improve generalization. DFQ leverages channel-wise decoupled deep features as queries, guided by cross-attention mechanisms, to learn a more expressive and generalized representation. A relaxed deep whitening transformation is introduced to decorrelate features, reducing channel redundancy and enhancing representation ability. The proposed method is integrated into Transformer segmentation models and demonstrates state-of-the-art performance on two benchmarks: optic cup/disk segmentation on fundus images and prostate segmentation on MRI. Extensive experiments show that the proposed method outperforms existing methods by at least 1.31% and 1.98% in Dice Similarity Coefficient (DSC) on the Fundus and Prostate benchmarks, respectively. The paper also includes ablation studies and visualized segmentation results to validate the effectiveness of the proposed approach.