Long-text caption generation for surgical image with a concept retrieval augmented large multimodal model | Synapse