ISSN 1004-4140
CN 11-3017/P

CT窗口设置对人工智能分类肺部病变结果的影响

程晓悦, 吴晓华, 郝优, 贺文, 李华, 刘佳宝, 曹邱婷

程晓悦, 吴晓华, 郝优, 等. CT窗口设置对人工智能分类肺部病变结果的影响[J]. CT理论与应用研究, 2023, 32(4): 515-522. DOI: 10.15953/j.ctta.2022.210.
引用本文: 程晓悦, 吴晓华, 郝优, 等. CT窗口设置对人工智能分类肺部病变结果的影响[J]. CT理论与应用研究, 2023, 32(4): 515-522. DOI: 10.15953/j.ctta.2022.210.
CHENG X Y, WU X H, HAO Y, et al. Effect of Computed Tomography Window Technique on the Results of Artificial Intelligence Classification of Lung Lesions[J]. CT Theory and Applications, 2023, 32(4): 515-522. DOI: 10.15953/j.ctta.2022.210. (in Chinese).
Citation: CHENG X Y, WU X H, HAO Y, et al. Effect of Computed Tomography Window Technique on the Results of Artificial Intelligence Classification of Lung Lesions[J]. CT Theory and Applications, 2023, 32(4): 515-522. DOI: 10.15953/j.ctta.2022.210. (in Chinese).

CT窗口设置对人工智能分类肺部病变结果的影响

基金项目: 国家中医药管理局中医药创新团队及人才支持计划(组织液循环网络(ZYYCXTD-D-202202));国家重点研发计划(复杂时变场景的物理仿真关键技术(2017YFB1002700));国家973项目(疏松结缔组织中的传递现象及艾灸等效应的模式化研究(2015CB554507));国家自然科学基金(图像不变性分析与应用(61379082));基于CT影像的肝脏肿瘤冷冻消融热传导数值模型与手术规划算法研究(6227012006))。
详细信息
    作者简介:

    程晓悦: 女,博士,首都医科大学附属北京友谊医院放射科主治医师,主要从事影像诊断工作,E-mail:cxiaotu@163.com

    吴晓华: 女,首都医科大学附属北京友谊医院放射科副主任医师,中国防痨协会多学科诊疗专业分会委员、北京医学会放射学分会感染学组委员,主要从事胸部影像诊断工作,E-mail:luckyemilyxh@aliyun.com

    通讯作者:

    吴晓华: 女,首都医科大学附属北京友谊医院放射科副主任医师,中国防痨协会多学科诊疗专业分会委员、北京医学会放射学分会感染学组委员,主要从事胸部影像诊断工作,E-mail:luckyemilyxh@aliyun.com

  • 中图分类号: O  242;TP  391;R  814

Effect of Computed Tomography Window Technique on the Results of Artificial Intelligence Classification of Lung Lesions

  • 摘要: 目的:应用3种不同的3D CNN的算法及5种CT窗口设置,探讨CT窗技术对人工智能分类肺部病变结果的影响。方法:回顾性分析172例周围型肺癌及185例局灶性肺炎的胸部CT影像资料,选择ResNet、ResNext以及DenseNet 3种不同的3D CNN的算法将病变分为两组,并在每1种3D CNN算法处理过程中应用5种不同的CT窗口设置,包括肺窗(1500,-600),纵隔窗(350,40),自定义窗口1(SW1)(1000,40),自定义窗口2(SW2)(1000,-100),全窗(4096,1024),分别计算分类准确率及AUC结果,并进行ROC曲线的两两对比。结果:ResNet的平均分类准确率最低为纵隔窗85.732%,AUC值为0.871;平均分类准确率最高为全窗,达91.596%,AUC值为0.946。ResNext的平均分类准确率最低为纵隔窗81.528%,AUC值为0.814;平均分类准确率最高为全窗,达86.568%,AUC值为0.882。DenseNet的平均分类准确率最低为纵隔窗87.954%,AUC值为0.906;平均分类准确率最高为SW2,达93.274%,AUC值为0.951。应用medcalc将3种3D CNN的5种窗口下的ROC曲线进行了两两对比发现,纵隔窗与肺窗、纵隔窗与SW1、纵隔窗与SW2之间的AUC值比较均有统计学意义。结论:3种3D CNN的分类诊断效能差别不大;CT窗口设置对CNN分类肺部病变结果有影响,在纵隔窗设置下以上3种人工智能算法对该两类肺部病变的诊断效能最差。
    Abstract: Objective: To use three different 3D CNN algorithms and five different computed tomography (CT) window settings to study the effect on the results of artificial intelligence classification of lung lesions in different CT window techniques. Method: A total of 172 cases of peripheral lung cancer and 185 of focal pneumonia who underwent chest CT were analyzed. Three different 3D CNN algorithms were selected (ResNet, ResNext, and DenseNet) to divide the lesions into two groups. Five different CT window settings, including lung window (1500, 600), mediastinal window (350, 40), custom window 1 (SW1) (1000, 40), and custom window 2 (SW2) (1000, 100), were used retrospectively. We calculated classification accuracy, receiver operating characteristic (ROC) curve, and area under the curve (AUC). The ROC curve was compared in pairs. Results: The average classification accuracy of ResNet was the lowest in the mediastinal window (85.732%; AUC value: 0.871) and the highest in the full window (91.596%; AUC value: 0.946). The average classification accuracy of ResNext was the lowest in the mediastinal window (81.528%; AUC value: 0.814) and the highest in the full window (86.568%; AUC value: 0.882). The average classification accuracy of DenseNet was the lowest in the mediastinal window (87.954%; AUC value: 0.906) and the highest in the SW2 window (93.274%; AUC value: 0.951). Medcalc was used to compare ROC curves under five windows of three 3D CNN. The AUC values between mediastinal window and lung window, mediastinal window and SW1, and mediastinal window and SW2 were statistically significant. Conclusion: There is little difference in the diagnostic efficacy of the three 3D CNN. Different CT window settings have an influence on the results of CNN classification of the lung lesions, and the diagnostic efficiency of the three 3D CNN is the worst under the mediastinal window.
  • 图  1   不同窗口设置时对病灶的显示不同:(a)~(e)左肺上叶周围型肺癌(腺癌);(f)~(j)右肺下叶局灶性肺炎

    (a)和(f)肺窗;(b)和(g)纵隔窗;(c)和(h)SW1;(d)和(i)SW2;(e)和(j)全窗

    Figure  1.   Different display of lesions in different windows: (a)~(e) is the same patient, diagnosed as peripheral lung cancer in the upper lobe of the left lung (pathological examination shows adenocarcinoma); (f)~(j) is the same patient, diagnosed as focal pneumonia in the lower lobe of the right lung

    图  2   3D ResNet网络构建

    主干使用18层的3D ResNet网络,在每个卷积块中,有1个卷积层(其参数在图中列出)、1个批标准化(Batch Normalization)层和1个ReLu激活层,图中的第1个参数“3×3×3”表示3D内核大小,第3个参数“64、128、256、512”表示通道数,最后1个参数“/2”表示步长为2的池层。

    Figure  2.   3D ResNet network design

    图  3   (a)ResNet的一个构建块;(b)基数为32的ResNext构建块[7]

    Figure  3.   (a) A building block of ResNet; (b) ResNext building block with 32 cardinality[7]

    图  4   DenseNet连接示意图[8]

    Figure  4.   DenseNet connection diagram[8]

    图  5   ResNet 5个窗口设置的ROC曲线图

    Figure  5.   ROC curve of ResNet

    图  6   ResNext 5个窗口设置的ROC曲线图

    Figure  6.   ROC curve of ResNext

    图  7   DenseNet 5个窗口设置的ROC曲线图

    Figure  7.   ROC curve of DenseNet

    表  1   ResNet 5个窗口设置的5倍交叉验证的分类准确率(%)及AUC结果

    Table  1   Classification accuracy (%) and AUC results of 5-fold cross validation of 5 window settings in ResNet

    窗口(WW,WL)Fold 1Fold 2Fold 3Fold 4Fold 5平均分类准确率/%AUC
      肺窗 (1500,−600)87.5088.7391.5591.5593.0690.4780.929
      纵隔窗 (350,40)80.5687.3290.1485.9284.7285.7320.871
      SW1 (1000,40)86.1192.9688.7391.5590.2889.9260.918
      SW2 (1000,−100)87.5092.9690.1491.5590.2890.4860.923
      Full (4096,1024)87.5090.1490.1494.3795.8391.5960.946
    下载: 导出CSV

    表  2   ResNext 5个窗口设置的5倍交叉验证的分类准确率(%)及AUC结果

    Table  2   Classification accuracy (%) and AUC results of 5-fold cross validation of 5 window settings in ResNext

    窗口(WW,WL)Fold 1Fold 2Fold 3Fold 4Fold 5平均分类准确率/%AUC
      肺窗 (1500,−600)84.7280.1487.3288.7387.5085.6820.896
      纵隔窗 (350,40)77.7883.1081.6984.5180.5681.5280.814
      SW1 (1000,40)86.1185.9288.7385.9284.7286.2800.881
      SW2 (1000,−100)84.7288.7383.1085.9287.5085.9940.878
      Full (4096,1024)83.3391.5585.9287.3284.7286.5680.882
    下载: 导出CSV

    表  3   DenseNet 5个窗口设置的5倍交叉验证的分类准确率(%)及AUC结果

    Table  3   Classification accuracy (%) and AUC results of 5-fold cross validation of 5 window settings in DenseNet

    窗口(WW,WL)Fold 1Fold 2Fold 3Fold 4Fold 5平均分类准确率/%AUC
      肺窗 (1500,−600)87.5091.5592.9694.3793.0691.8880.944
      纵隔窗 (350,40)88.8987.3291.5584.5187.5087.9540.906
      SW1 (1000,40)94.4494.3790.1490.1493.0692.4300.932
      SW2 (1000,−100)93.0695.7791.5591.5594.4493.2740.951
      Full (4096,1024)91.6792.9690.1494.3794.4492.7160.941
    下载: 导出CSV
  • [1]

    Van GINNEKEN B. Fifty years of computer analysis in chest imaging: Rule-based, machine learning, deep learning[J]. Radiological Physics and Technology, 2017, 10(1): 23−32. doi: 10.1007/s12194-017-0394-5

    [2]

    YU X, LU S, GUO L, et al. ResGNet-C: A graph convolutional neural network for detection of COVID-19[J]. Neurocomputing, 2021, 452: 592−605. doi: 10.1016/j.neucom.2020.07.144

    [3]

    LEE S M, SEO J B, YUN J, et al. Deep learning applications in chest radiography and computed tomography: Current state of the art[J]. Journal of Thoracic Imaging, 2019, 34(2): 75−85. doi: 10.1097/RTI.0000000000000387

    [4]

    SHIN H C, ROTH H R, GAO M, et al. Deep convolutional neural networks for computer-aided detection: CNN architectures, dataset characteristics and transfer learning[J]. IEEE Transactions on Medical Imaging, 2016, 35(5): 1285−1298.

    [5]

    KANG G, LIU K, HOU B, et al. 3D multi-view convolutional neural networks for lung nodule classification[J]. Plos One, 2017, 12(11): 12−22.

    [6]

    CHENG X, WEN H, YOU H, et al. Recognition of peripheral lung cancer and focal pneumonia on chest computed tomography images based on convolutional neural network[J]. Technology in Cancer Research & Treatment, 2022, 21: 1-12.

    [7]

    XIE S, GIRSHICK R, DOLLÁR P, et al. Aggregated residual transformations for deep neural networks[C]//In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2017: 1492-1500.

    [8]

    HUANG G, LIU Z, Van der MAATEN L, et al. Densely connected convolutional networks[C]//In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, 2017: 4700-4708.

    [9] 田培林, 徐覃莎, 唐利荣, 等. 胸部CT检查的肺窗技术[J]. 生物医学工程与临床, 2002,6(4): 209−213. doi: 10.3969/j.issn.1009-7090.2002.04.010

    TIAN P L, XU Q S, TANG L R, et al. Window techniques for pulmonary image in chest CT[J]. Biomedical Engineering and Clinical Medicine, 2002, 6(4): 209−213. (in Chinese). doi: 10.3969/j.issn.1009-7090.2002.04.010

    [10] 张胜超, 陈浩, 秦宣, 等. CT窗口技术在肺磨玻璃结节诊断中的临床意义[J]. 中国现代医学杂志, 2019,29(14): 106−109.

    ZHANG S C, CHEN H, QIN X, et al. Clinical significance of CT window technique in the diagnosis and treatment of pulmonary ground-glass nodules[J]. China Journal of Modern Medicine, 2019, 29(14): 106−109. (in Chinese).

    [11]

    KITAMI A, SANO F, HAYASHI S, et al. Correlation between histological invasiveness and the computed tomography value in pure ground-glass nodules[J]. Surgery Today, 2016, 46(5): 593−598. doi: 10.1007/s00595-015-1208-1

    [12] 张鹏, 徐欣楠, 王洪伟, 等. 基于深度学习的计算机辅助肺癌诊断方法[J]. 计算机辅助设计与图形学学报, 2018,30(1): 90−99.

    ZHANG P, XU X N, WANG H W, et al. Computer-aided lung cancer diagnosis approaches base on deep learning[J]. Journal of Computer-Aided Design & Computer Graphics, 2018, 30(1): 90−99. (in Chinese).

    [13] 尹柯, 张久权, 伍建林, 等. 对比卷积神经网络分类模型与放射科医师鉴别浸润性肺腺癌的效能[J]. 中国医学影像技术, 2021,37(9): 1338−1342.

    YIN K, ZHANG J Q, WU J L, et al. Comparison on convolutional neural network classification model and radiologists in differentiating invasive lung adenocarcinoma[J]. Chinese Journal of Medical Imaging Technology, 2021, 37(9): 1338−1342. (in Chinese).

    [14] 戴书华, 刘国芳, 向东生. 肺磨玻璃结节CT值测量在早期癌症诊断中的意义[J]. 中华肺部疾病杂志(电子版), 2019,12(6): 770−771.

    DAI S H, LIU G F, XIANG D S. The significance of CT value measurement of pulmonary ground glass nodule in early cancer diagnosis[J]. Chinese Journal of Lung Diseases Electronic Edition, 2019, 12(6): 770−771. (in Chinese).

    [15]

    HE S, CHEN C, WANG Z, et al. The use of the mean computed tomography value to predict the invasiveness of ground-glass nodules: A meta-analysis[J]. Asian Journal of Surgery, 2022, 18: 1-6.

图(7)  /  表(3)
计量
  • 文章访问数:  422
  • HTML全文浏览量:  208
  • PDF下载量:  29
  • 被引次数: 0
出版历程
  • 收稿日期:  2022-10-28
  • 修回日期:  2022-12-05
  • 录用日期:  2022-12-06
  • 网络出版日期:  2023-01-03
  • 发布日期:  2023-07-30

目录

    /

    返回文章
    返回
    x 关闭 永久关闭