Segment anything models (SAMs) demonstrate exceptional zero-shot segmentation capabilities on natural images when provided appropriate prompts. However, directly applying SAMs to medical images presents challenges due to the complexity and diversity of m