Lesions in medical images tend to exhibit characteristics such as tiny size, heterogeneous morphology, and diffuse boundaries, leading to low-contrast fusion with background tissues, which puts higher requirements on the accuracy and generalization ability of segmentation algorithms. In this paper, Wavelet-Context-Gated Network (WCG-Net) is proposed for achieving accurate segmentation of complex lesions. First, the Wavelet-Enhanced NeXt-Generation Multi-Frequency Fusion Block (WaveNeXt Block) module is designed for lesion morphology modeling, which constructs global morphological representations in the low-frequency domain through the wavelet frequency splitting mechanism, and at the same time reinforces the edge texture details in the high-frequency domain to achieve global context-guided structure-boundary fusion. Second, to parse the irregular edge features of the lesion, the Wavelet-Augmented Context-Aware Attention (WA-CAA) module is designed to achieve multi-scale spatial calibration through encoder-decoder cross-layer feature fusion, which combines the spatial attention to dynamically enhance the center and edge regions with remote dependency. Finally, to improve the model segmentation stability and generalization ability, the Gated Channel Transformation (GCT) module is introduced to explicitly model the inter-channel competition-collaboration relationship using learnable gating variables, which dynamically suppresses redundant features and strengthens the lesion-sensitive response. Our method achieves superior segmentation accuracy on BUSI, GlaS, and MoNuSeg datasets compared to existing methods.