深度学习算法与影像医师测量肺转移瘤体积的一致性分析  

Consistency analysis between deep learning algorithm and radiologists in measuring the volume of lung metastases

作　　者：龙创 周颖俊 Long Chuang;Zhou Yingjun(Department of Radiology,Xiangtan Central Hospital,Xiangtan 411100,China)

机构地区：[1]湘潭市中心医院放射科,湘潭411100

出　　处：《现代仪器与医疗》2024年第4期97-100,共4页Modern Instruments & Medical Treatment

基　　金：湘潭市医学会项目(2023xtyx-46)。

摘　　要：目的 评估深度学习(Deep learning,DL)算法与放射科医师在测量肺转移瘤体积方面的一致性。方法 从湘潭市中心医院2019年6月—2023年6月收治的肿瘤患者中随机选取57例肺转移瘤患者的CT扫描图像,其中包括89个实性转移结节。使用一种商业DL算法自动识别肺转移结节,并通过DL计算肺转移瘤体积(Lung metastases volume by DL,LMV-DL)。同时,两名资深放射科医师在肺窗上手动勾画肺转移结节,并使用面积求和法计算LMV-Radiologist1和LMV-Radiologist2。然后,使用Bland-Altman方法计算三组LMV之间的95%一致性界限(95%LoA),并评估一致性。结果 Bland-Altman法示:三组两两比较,其中,LMV-DL与LMV-Radiologist1、LMV-Radiologist2的95%LoA分别为-758.3~416.2mm^(3)、-627.1~518.1mm^(3),宽于Radiologist1与Radiologist2之间95%LoA-207.1~440.2mm^(3)。结论 DL算法在测量LMV方面与放射科医师表现出良好的一致性,并可作为自动测量方法替代手动测量,有助于肺结节的临床管理。Objective To assess the agreement between a deep learning(DL) algorithm and radiologists in measuring lung metastases volume.Methods CT scan images of 57 patients with lung metastases were randomly selected from Xiangtan Central Hospital from June 2019 to June 2023,including 89 solid metastatic nodules.A commercial DL algorithm was used to automatically identify pulmonary metastatic nodules and calculate lung metastases volume by DL(LMV-DL).Simultaneously,two senior radiologists manually outlined the lung metastatic nodules on the lung window and calculated LMV-Radiologist 1and LMV-Radiologist 2 using the area summation method.The Bland-Altman method was then used to calculate the 95% limits of agreement(95% LoA) between the three groups of LMV,and the agreement was assessed.Results The Bland-Altman method showed that the 95% LoA between the DL algorithm and each radiologist([-758.3 to 416.2] mm^(3) and [-627.1 to 518.1] mm^(3),respectively) was wider than the pairwise comparisons between radiologists(-207.1 to 440.2) mm^(3).Conclusion The DL algorithm demonstrated good consistency with radiologists in measuring LMV and can serve as an automatic measurement method to replace manual measurement,aiding in the clinical management of lung nodules.

关 键 词：深度学习算法 放射科医师 肺转移瘤 评估一致性 临床管理1 

分 类 号：TH77[机械工程—仪器科学与技术] R681.5[机械工程—精密仪器及机械]