<p>Deep learning-based medical image segmentation methods have achieved significant progress in recent years. However, medical ultrasound image segmentation remains challenging due to small lesion regions, blurred boundaries, low contrast, and the loss of local details. To address these challenges, this paper proposes a coarse-to-fine two-stage collaborative medical image segmentation network based on memristive neural networks, named the memristive coarse-to-fine segmentation network (MCFSNet). The proposed MCFSNet consists of a coarse segmentation sub-network (MCS-Net) and a coarse segmentation result-guided dual-encoder single-decoder refinement network (RS-Net), forming a progressive framework from coarse localization to refined segmentation. MCS-Net performs initial lesion localization through an encoder-decoder structure and maps the multiply-accumulate operations in convolutional layers onto memristor crossbar arrays to improve computational efficiency. The generated coarse segmentation result provides spatial prior information for the subsequent refined segmentation stage. Guided by the coarse segmentation result, RS-Net extracts local edge features and global semantic features through a dual-encoder structure, and integrates them through the memristive cross-attention module (MCAM) and the dual-branch memristive feature interaction fusion module (DBMF). Specifically, MCAM establishes correlations between local edge features and global semantic information, thereby alleviating blurred boundaries and missed segmentation of small lesions. DBMF adaptively enhances the fusion of edge detail features and global semantic features, improving the structural completeness and boundary clarity of the segmentation results. Experimental results on the BUSI and DDTI datasets demonstrate that the proposed method achieves favorable segmentation performance, verifying the effectiveness of the proposed network architecture.</p>