CT影像组学鉴别乳糜性肺炎和感染性肺炎的价值  

The Value of CT Radiomics in Differentiating Chylous Pneumonia from Infectious Pneumonia

作　　者：张妍 郝琪 李兴鹏 安亚慧 刘梦珂 徐井旭 黄陈翠 孙小丽 王仁贵 ZHANG Yan;HAO Qi;LI Xingpeng(Department of Radiology,Beijing Shijitan Hospital,Capital Medical University,Beijing 100038,P.R.China)

机构地区：[1]首都医科大学附属北京世纪坛医院放射科,100038 [2]北京大学第九临床医学院放射科,100038 [3]北京深睿博联科技有限责任公司,100089

出　　处：《临床放射学杂志》2023年第11期1762-1768,共7页Journal of Clinical Radiology

基　　金：国家自然科学基金项目(编号:61876216);首都医科大学附属北京世纪坛医院项目(编号:2020-q18);北京市科学技术委员会基金项目(编号:Z211100003521009)。

摘　　要：目的评估CT影像组学模型对乳糜性肺炎和感染性肺炎的鉴别诊断价值。方法搜集确诊的乳糜性肺炎和感染性肺炎患者各83例,按照训练集与验证集7∶3的比例随机拆分,训练集共纳入116例患者,验证集共纳入50例患者。由放射科医师应用深睿多模态科研平台对患者的感兴趣区进行标注,提取影像组学特征2107个。采用相关性分析及F Test算法筛选特征,采用Logistic回归分类器构建影像组学模型,绘制受试者工作特征曲线,并将模型的预测结果与3位不同年资医师的诊断结果比较。结果共筛选出10个对鉴别乳糜性肺炎和感染性肺炎最具有价值的特征用于建模。训练集和验证集中影像组学模型鉴别乳糜性肺炎和感染性肺炎的曲线下面积(95%CI)分别为0.91(0.85~0.96)和0.88(0.78~0.98),敏感度分别为0.86和0.88,特异度分别为0.88和0.76,准确度分别为0.87和0.82。3位医师的诊断乳糜性肺炎和感染性肺炎的曲线下面积(95%CI)分别为0.58(0.48~0.70)、0.74(0.66~0.86)、0.86(0.80~0.95),敏感度分别为0.52、0.72、0.84,特异度分别为0.64、0.76、0.88,准确度分别为0.58、0.74、0.86。影像组学模型与初级医师的诊断效能差异有统计学意义(P<0.001),与中级和高级医师差异均无统计学意义(P=0.09,P=0.78),初级与中级医师诊断乳糜性肺炎的差异有统计学意义(P=0.002),初级与高级医师诊断乳糜性肺炎的差异有统计学意义(P<0.001),中级与高级医师诊断乳糜性肺炎的差异有统计学意义(P=0.02)。结论基于胸部CT平扫的影像组学模型可有效鉴别乳糜性肺炎和感染性肺炎,此模型的预测结果与影像科中、高年资医师的诊断结果无显著性差异。Objective To evaluate the value of CT radiomics model in the dfferential diagnosis of chylous pneumonia and infectious pneumonia.Methods A total of 83 patients with chylous pneumonia and 83 patients with infectious pneumonia diagnosed were randomly divided into training set and validation set according to the ratio of 7:3.116 patients were included in the training set and 50 patients were included in the validation set.The region of interest of patients was marked by radiologists using the Deepwise multimodal scientific research platform.A total of 2107 radiomics features were extracted.Correlation analysis and F Test algorithm were used to screen features.The Logistic regression classifier was used to construct the radiomics model,and the receiver operating characteristic curve was drawn.The prediction results of the model were compared with the diagnosis results of three doctors with different seniority.ResultsA total of 10 most valuable features for differentiating chylous pneumonia from infectious pneumonia were selected for modeling.The area under the curve(95%CI)of the radiomics model in the training set and the validation set was 0.91(0.85-0.96)and 0.88(0.78-0.98),the sensitivity was 0.86 and 0.88,and the specificity was 0.88 and 0.76,respectively.The accuracy was 0.87 and 0.82,respectively.The area under the curve(95%CI)of the three doctors for the diagnosis of chylous pneumonia and infectious pneumonia were 0.58(0.48-0.70),0.74(0.66-0.86),0.86(0.80-0.95),respectively,and the sensitivity were 0.52,0.72,0.84,respectively.The specificity were 0.64,0.76,0.88,and the accuracy were 0.58,0.74,0.86,respectively.The diagnostic performance of the radiomics model was significantly different from that of junior doctors(P<0.001),and there was no significant difference between junior and senior doctors(P=0.09,P=0.78).There was a significant difference between junior doctors and intermediate doctors in the diagnosis of chylous pneumonia(P=0.002).There was a statistically significant difference in the diagnosis of chylo

关 键 词：影像组学 乳糜性肺炎 磨玻璃影 体层摄影术 X线计算机 

分 类 号：R816.4[医药卫生—放射医学] R563.1[医药卫生—临床医学]