Structured Prompt-Guided Segmentation of Enhanced CT Images for Predicting Lateral Cervical Lymph Node Metastasis in Papillary Thyroid Carcinoma

Accurate segmentation of metastatic lymph nodes in head-and-neck CT is crucial for treatment planning in papillary thyroid carcinoma. Although deep networks have advanced medical image segmentation, most existing methods depend solely on visual features and neglect clinical semantic cues. In this work, we propose a text-guided segmentation framework that introduces structured clinical baseline feature into a YOLOv8-based segmentation network. To effectively fuse semantic and spatial features, we design a cross-modal fusion mechanism that aligns textual embeddings with image features at multiple scales. This allows the model to focus on diagnostically relevant regions and improves performance in challenging scenarios involving low contrast and ambiguous boundaries. Extensive experiments on a multi-center, pathology-confirmed CT dataset demonstrate that our method achieves superior results in AP@50, mAP@[.5:.95], and AR@[.5:.95], outperforming existing state-of-the-art instance segmentation models.