多模态生成式人工智能模型的温度设置与人工智能生成胸部X线报告准确性和质量的关联
Temperature Setting of a Multimodal Generative Artificial Intelligence Model: Association With Accuracy and Quality of Artificial Intelligence-Generated Chest Radiograph Reports.
PubMed 文献 · AJR Am J Roentgenol · 2026-02-25
该研究评估多模态生成式人工智能模型在17种温度设置(0.0至1.6)下生成胸部X线报告的表现。较低温度与报告准确性、可读性、放射科医生可接受度和主观质量改善相关,且幻觉更少。结果强调该参数在潜在临床部署中的重要性。
要点
- 较低温度设置与人工智能生成胸部X线报告的准确性改善相关。
- 较低温度设置与报告可读性、放射科医生可接受度和主观质量改善相关。
- 较低温度设置与幻觉减少相关。
核心结论
- 较低温度设置与人工智能生成胸部X线报告的准确性、可读性、放射科医生可接受度和主观质量改善相关,且幻觉更少。
- 结果强调温度参数在潜在临床部署中的重要性。
研究设计与样本
- 研究评估多模态生成式人工智能模型在17种温度设置(0.0至1.6)下生成胸部X线报告的表现。
局限性
- 原文未报告局限性。
以上解读依据论文摘要(全文未开放获取或许可不允许翻译转载)。
英文原文(摘要)· 点击展开
A multimodal generative artificial intelligence model created radiology reports for chest radiographs across 17 temperature settings (from 0.0 to 1.6). Lower temperatures were associated with improved report accuracy, readability, radiologist acceptability, and subjective quality, with fewer hallucinations. These results highlight this parameter’s importance during potential clinical deployment.
来源:AJR Am J Roentgenol,查看原文。版权归原作者所有。
引用
- Hong EK, Lee S, Song O, et al. Temperature Setting of a Multimodal Generative Artificial Intelligence Model: Association With Accuracy and Quality of Artificial Intelligence-Generated Chest Radiograph Reports. AJR Am J Roentgenol. 2026. PMID: 41159788. doi:10.2214/AJR.25.33704
供执业医师工作参考,不替代临床判断;数值与分级以指南原文为准。