<p>Computed tomography (CT) plays a crucial role in differentiating and diagnosing kidney tumors, and accurate tumor detection in CT images can significantly aid physicians in making preliminary diagnoses. To address the challenges posed by tumor scale diversity and edge ambiguity in CT images, we propose a Dual-Stream Edge-Aware Network (DSEA-Net) for kidney tumor segmentation. DSEA-Net employs a subject-edge feature fusion mechanism with two decoder feature streams: one focusing on the tumor subject and the other on its edges. Specifically, we introduce the Parallel Residual Feature Enhancer in the tumor subject feature decoder, which enhances segmentation performance by dynamically optimizing and selectively refining multi-scale tumor features. Additionally, in the tumor edge feature decoder, we design the Pyramid Edge-Aware Decoder, which extracts and fuses multi-scale edge features, effectively improving the network’s sensitivity to edge information. Finally, we propose the Double Interactive Fusion block, which deeply fuses the tumor subject and edge feature streams, enabling the segmentation network to leverage both features synergistically for improved tumor segmentation. We validated our model on a constructed dataset comprising 451 contrast-enhanced abdominal CT scans, and experimental results demonstrate the effectiveness and superiority of DSEA-Net over existing methods.</p>