ChestXLLaMA v1.0: Chain-of-Thought Annotations, Prompts, and Generated Reports for Chest X-ray Reasoning

This dataset accompanies the paper “ChestXLLaMA: A Chain-of-Thought-Enhanced Vision–Language Model for Clinical Reasoning in Chest X-ray Reporting.”
It contains the chain-of-thought (CoT) prompts, teacher-model instructions, and example annotated reports used for training and evaluating the ChestXLLaMA model. Each sample pairs a reference chest X-ray report with a ten-step radiology reasoning sequence generated by expert teacher models. No patient images or protected health information are included; only textual annotations, prompts, and generated reasoning outputs. The data support reproducibility of the model’s reasoning-enhanced training process.

本数据集配套论文《ChestXLLaMA：一款面向胸部X射线报告临床推理的思维链增强型视觉语言模型》。其包含用于训练与评估ChestXLLaMA模型的思维链（Chain-of-Thought, CoT）提示词、教师模型指令集及经标注的示例报告。每个样本均将参考胸部X射线报告与专家教师模型生成的十步放射学推理序列进行配对。数据集中未包含患者影像或受保护健康信息，仅涵盖文本标注、提示词及生成的推理输出。该数据集可为该模型的推理增强训练流程的可复现性提供支撑。