亚实性肺结节CT阈值分割：实性成分识别与定量
发布时间：2018-11-23 12:58
【摘要】：背景与目的在胸部计算机断层扫描(computed tomography,CT)图像上肺内亚实性结节(subsolid nodule,SSN)是指纯磨玻璃结节和部分实性结节。SSN实性成分的识别与定量对鉴别诊断,预测病理和评估预后具有重要价值,但目前缺乏公认且客观的标准。对亚实性结节内实性成分做CT体积定量的研究报告尚少。本研究旨在探究CT阈值分割法判断SSN类型并定量其实性成分体积的阈值。方法共纳入102例SSN。由观察者1和观察者2分别独立对结节内有无实性成分,即结节的类型(部分实性或纯磨玻璃)进行主观判断,结果不一致时采纳观察者3的意见,由此确定出所有结节的类型并以此作为评估阈值分割法判断SSN类型效能的参照标准。被判定为部分实性的结节由观察者1和观察者2分别独立对其实性成分进行体积测量,测量时借助Auto Contour软件包并辅以手动调整。以两位观察者所得实性成分体积的平均值作为评估阈值分割法定量SSN实性成分体积的参照标准。由观察者1对全部结节进行阈值分割,步骤为首先采用Auto Contour软件包对结节进行整体提取并记录结节的整体体积,然后使用3D-color-ROI工具计算所获结节内不同CT值区间的体素体积。共设9个CT值区间,下限值(阈值)分别设为-500 HU、-450 HU、-400HU、-350 HU、-300HU、-250 HU、-200 HU、-160 HU、-130 HU,上限值均设为2000 HU,假定上述CT值区间的体素均为实性成分。计算阈值分割所获实性成分体积与结节整体体积的比率(%)。以观察者确定的结节类型为状态变量,以体积比率为检验变量,绘制不同CT值区间判断结节类型的受试者工作特征(receiver operating characteristic,ROC)曲线,得到曲线下面积(area under curve,AUC)。应用DeLong检验筛选CT阈值分割判断SSN类型的阈值。通过最大Youden指数得到确认结节存在实性成分的体积比率界限值。对阈值分割所得体积与实性成分体积参照标准之间进行配对Wilcoxon检验,筛选可用于实性成分体积定量的阈值。结果阈值为-250 HU时判断亚实性结节类型的准确度最高(AUC=0.982),对应体积比率界限值为1.10%,此时敏感度、特异度分别为100.0%、89.7%;阈值为-300 HU时判断亚实性结节类型的准确度次之(AUC=0.977),对应体积比率界限值为6.14%,此时敏感度、特异度分别为90.5%、94.9%。然而阈值-250 HU与阈值-350 HU、-300 HU、-200 HU、-160 HU、-130 HU在判断亚实性结节类型上差异不显著(P均0.05)。阈值为-250 HU、-300 HU时所得实性成分体积202.7mm~3(598.2 mm~3)、247.1 mm~3(696.0 mm~3)与参照标准199.5 mm~3(743.1 mm~3)间无显著差异(P=0.1251、0.0613),而其他阈值所得实性成分体积均与参照标准间存在显著差异(P均0.05)。结论本研究表明,CT阈值分割能够可靠地对SSN的类型进行判断并对其实性成分体积进行定量评估;阈值可设为-250 HU或-300 HU。
[Abstract]:Background & objective on chest computed tomography (computed tomography,CT) images, subsolid pulmonary nodules (subsolid nodule,SSN) are pure ground glass nodules and partial solid nodules. It is important to predict pathology and evaluate prognosis, but there is a lack of accepted and objective criteria. There are few studies on CT volume quantification of solid components in subsolid nodules. The purpose of this study was to explore the threshold value of CT threshold segmentation to determine the SSN type and to quantify the volume of the actual component. Methods 102 cases of SSN. were included Observer 1 and Observer 2 made independent subjective judgments on whether there were solid components in the nodules, that is, the types of nodules (partially solid or pure ground glass). When the results were inconsistent, the opinion of Observer 3 was adopted. The types of all nodules are determined and used as a reference criterion for evaluating the effectiveness of SSN type by threshold segmentation method. The nodules determined to be partially solid were measured independently by Observer 1 and Observer 2, respectively, with the help of Auto Contour software package and manual adjustment. The mean value of real component volume obtained by two observers is used as the reference criterion for evaluating the volume of SSN real component by threshold segmentation method. The threshold value of all the nodules was segmented by observer 1. The steps were as follows: firstly, the whole nodules were extracted by Auto Contour software package and the whole volume of the nodules was recorded. Then the volume of voxel in different CT values of the nodules was calculated by using the 3D-color-ROI tool. There are 9 CT ranges, and the lower limit (threshold) is -500 HU,-450 HU,-400HU,-350 HU,-300HU,-250 HU,-200 HU,-160 HU,-130 HU, and the upper limit is 2000 HU,. It is assumed that the voxels in the above CT interval are real components. Calculate the ratio of the solid component volume to the whole nodule volume obtained by threshold segmentation (%). Taking the nodular type determined by the observer as the state variable and the volume ratio as the test variable, the (receiver operating characteristic,ROC curves with different CT values to judge the nodule type were drawn, and the area under the curve (area under curve,AUC) was obtained. DeLong test was used to select the threshold value of CT to judge the threshold of SSN type. By using the maximum Youden exponent, the boundary value of the volume ratio is obtained to confirm the existence of solid components in the nodules. The matched Wilcoxon test was performed between the volume of real component and the reference standard of real component volume, and the threshold value for quantitative quantification of real component volume was screened. Results when the threshold was -250 HU, the accuracy (AUC=0.982) of subsolid nodules was the highest, and the threshold value of the corresponding volume ratio was 1.10. The sensitivity and specificity were 100.0 and 89.7, respectively. When the threshold was -300 HU, the accuracy (AUC=0.977) of subsolid nodules was the second, and the corresponding threshold value of volume ratio was 6.14. At this time, the sensitivity and specificity were 90.5 and 94.9, respectively. However, there was no significant difference between the threshold of-250 HU and the threshold of-350 HU,-300 HU,-200 HU,-160 HU,-130 HU in the classification of subsolid nodules. There was no significant difference between 202.7mm~3 (598.2 mm~3), 247.1 mm~3 (696.0 mm~3) and 199.5 mm~3 (743.1 mm~3) when the threshold was -250 HU,-300 HU. However, the volume of solid components obtained from other thresholds was significantly different from that of reference standard (P 0.05). Conclusion this study shows that CT threshold segmentation can reliably judge the type of SSN and quantitatively evaluate the volume of actual components, and the threshold can be set to -250 HU or -300 HU..
【学位授予单位】：天津医科大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R734.2;R730.44
本文编号：2351704
[Abstract]:Background & objective on chest computed tomography (computed tomography,CT) images, subsolid pulmonary nodules (subsolid nodule,SSN) are pure ground glass nodules and partial solid nodules. It is important to predict pathology and evaluate prognosis, but there is a lack of accepted and objective criteria. There are few studies on CT volume quantification of solid components in subsolid nodules. The purpose of this study was to explore the threshold value of CT threshold segmentation to determine the SSN type and to quantify the volume of the actual component. Methods 102 cases of SSN. were included Observer 1 and Observer 2 made independent subjective judgments on whether there were solid components in the nodules, that is, the types of nodules (partially solid or pure ground glass). When the results were inconsistent, the opinion of Observer 3 was adopted. The types of all nodules are determined and used as a reference criterion for evaluating the effectiveness of SSN type by threshold segmentation method. The nodules determined to be partially solid were measured independently by Observer 1 and Observer 2, respectively, with the help of Auto Contour software package and manual adjustment. The mean value of real component volume obtained by two observers is used as the reference criterion for evaluating the volume of SSN real component by threshold segmentation method. The threshold value of all the nodules was segmented by observer 1. The steps were as follows: firstly, the whole nodules were extracted by Auto Contour software package and the whole volume of the nodules was recorded. Then the volume of voxel in different CT values of the nodules was calculated by using the 3D-color-ROI tool. There are 9 CT ranges, and the lower limit (threshold) is -500 HU,-450 HU,-400HU,-350 HU,-300HU,-250 HU,-200 HU,-160 HU,-130 HU, and the upper limit is 2000 HU,. It is assumed that the voxels in the above CT interval are real components. Calculate the ratio of the solid component volume to the whole nodule volume obtained by threshold segmentation (%). Taking the nodular type determined by the observer as the state variable and the volume ratio as the test variable, the (receiver operating characteristic,ROC curves with different CT values to judge the nodule type were drawn, and the area under the curve (area under curve,AUC) was obtained. DeLong test was used to select the threshold value of CT to judge the threshold of SSN type. By using the maximum Youden exponent, the boundary value of the volume ratio is obtained to confirm the existence of solid components in the nodules. The matched Wilcoxon test was performed between the volume of real component and the reference standard of real component volume, and the threshold value for quantitative quantification of real component volume was screened. Results when the threshold was -250 HU, the accuracy (AUC=0.982) of subsolid nodules was the highest, and the threshold value of the corresponding volume ratio was 1.10. The sensitivity and specificity were 100.0 and 89.7, respectively. When the threshold was -300 HU, the accuracy (AUC=0.977) of subsolid nodules was the second, and the corresponding threshold value of volume ratio was 6.14. At this time, the sensitivity and specificity were 90.5 and 94.9, respectively. However, there was no significant difference between the threshold of-250 HU and the threshold of-350 HU,-300 HU,-200 HU,-160 HU,-130 HU in the classification of subsolid nodules. There was no significant difference between 202.7mm~3 (598.2 mm~3), 247.1 mm~3 (696.0 mm~3) and 199.5 mm~3 (743.1 mm~3) when the threshold was -250 HU,-300 HU. However, the volume of solid components obtained from other thresholds was significantly different from that of reference standard (P 0.05). Conclusion this study shows that CT threshold segmentation can reliably judge the type of SSN and quantitatively evaluate the volume of actual components, and the threshold can be set to -250 HU or -300 HU..
【学位授予单位】：天津医科大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R734.2;R730.44
【参考文献】
相关期刊论文 前1条
1 王群;蒋伟;奚俊杰;;肺部多发磨玻璃影的外科治疗[J];中国肺癌杂志;2016年06期
，本文编号：2351704
本文链接：https://www.wllwen.com/yixuelunwen/fangshe/2351704.html
