Accurate medical image segmentation is often challenged by blurred boundaries caused by the high morphological and color similarity between lesions and surrounding tissue, thus compromising both segmentation accuracy and diagnostic reliability. To this end, this study proposes a fuzzy edge enhancement network based on residual differential convolution and designs a feature extraction module for edge awareness detail to enhance edge information via gradient feature coding. Meanwhile, the global edge detail bi-level routing attention is introduced in the bottleneck to integrate contextual information for precise modeling. Furthermore, a residual edge-aware localization module is constructed to enable refined localization of fuzzy edges. The experimental results show that on public colorectal cancer, skin lesion, and breast ultrasound images datasets, the proposed method significantly reduces the parameter count and computational complexity, with even stronger segmentation performance than existing state-of-the-art methods. Systematic ablation studies further validate the synergistic contribution of each component to edge detection, fully showcasing the potential application significance of the proposed method in improving the reliability of medical image analysis.