Accurate and efficient segmentation of skin lesions is fundamental to the advancement of computer-aided dermatological systems, where early and precise identification significantly influences clinical decision-making. Nonetheless, the creation of large-scale, pixel-level annotated datasets remains a significant bottleneck due to the expertise and time required for manual labeling. In this research, we introduce a semi-supervised segmentation framework that combines the strengths of U2-Net and the Segment Anything Model (SAM) to automate and refine the annotation process. Initially, U2-Net generates coarse lesion localization masks, which are subsequently enhanced through a prompt-based SAM refinement mechanism, producing high-quality pseudo-labels with minimal human intervention. This hybrid approach substantially reduces annotation costs while preserving, and in some cases enhancing, segmentation performance. Comprehensive experiments on the ISIC 2018 dataset reveal that our pipeline achieves competitive segmentation accuracy, offering a scalable and annotation-efficient solution that can serve as a foundation for subsequent lesion classification, progression monitoring, and broader dermatological image analysis tasks.