-
肺癌是在我国发病率和病死率最高的恶性肿瘤[1],其中,肺腺癌由于较早发生局部浸润和远处转移,其预后比肺鳞癌更差[2]。研究结果表明,微血管浸润状态可反映肿瘤的侵袭能力,是影响肺腺癌预后的重要因素,且美国国立综合癌症管理网络(NCCN)指南中也将血管浸润作为肺腺癌的高危险因素[2-5]。尽管术前鉴别肺腺癌是否发生微血管浸润对术后治疗策略的制定和预后有重要意义,但在实际临床工作中,传统影像学在判断肿瘤是否发生微血管浸润中的价值有限,甚至由于微血管浸润评估的工作量较大,国内大多数病理检查并未将微血管浸润作为常规检测项目。影像组学通过从医学图像中提取海量信息并进行处理,可获取传统影像学所不能反映的微观生物学信息,在疾病的诊断、疗效评价及预后评估等方面具有独特优势[6-9]。因此,有望通过影像组学对肺腺癌微血管浸润状态进行无创性预测。我们基于影像组学联合传统影像学征象建立综合模型,并验证该模型预测肺腺癌微血管浸润的效能。
-
患者基本信息及传统影像学征象见表1。其中,训练组中仅肿瘤最大径在微血管浸润阳性与阴性患者间的差异有统计学意义(t=5.580,P=0.035),其余的组间比较差异均无统计学意义。
项目 训练组(n=46) 验证组(n=19) 微血管浸润
阳性(n=21)微血管浸润
阴性(n=25)检验值 P值 微血管浸润
阳性(n=9)微血管浸润
阴性(n=10)检验值 P值 男性/女性(例) 11/10 8/17 χ2=1.955 0.162 5/4 9/1 χ2=2.898 0.238 年龄(岁) 62.90±12.57 59.36±8.09 t=4.574 0.254 60.78±12.97 59.20±7.93 t=4.169 0.750 肿瘤最大径(mm) 28.10±11.39 22.32±6.26 t=5.580 0.035 32.56±11.05 23.40±9.87 t=0.128 0.073 所在肺(左肺/右肺,例) 9/12 10/15 χ2=0.038 0.845 2/7 1/9 χ2=0.532 0.921 所在肺叶(上/中/下,例) 14/1/6 18/2/5 χ2=0.581 0.748 6/1/2 6/1/3 χ2=0.148 0.929 形状(类圆/不规则,例) 13/8 15/10 χ2=0.017 0.895 5/4 8/2 χ2=1.310 0.515 成分(实性/磨玻璃及混杂,例) 18/3 17/8 χ2=1.968 0.161 9/0 9/1 χ2=0.950 1.000 分叶征(有/无,例) 20/1 25/0 χ2=1.217 0.457 9/0 9/1 χ2=0.950 1.000 毛刺征(有/无,例) 10/11 17/8 χ2=1.955 0.162 6/3 8/2 χ2=0.434 0.891 空洞(有/无,例) 4/17 4/21 χ2=0.074 1.000 1/8 1/9 χ2=0.006 1.000 支气管充气征(有/无,例) 4/17 9/16 χ2=1.618 0.203 2/7 5/5 χ2=1.571 0.437 胸膜凹陷征(有/无,例) 15/6 20/5 χ2=0.461 0.497 7/2 8/2 χ2=0.014 1.000 血管集束征(有/无,例) 12/9 19/6 χ2=1.847 0.174 8/1 10/0 χ2=1.173 0.474 注:表中,HRCT:高分辨率CT 表 1 65例肺腺癌患者的临床资料及HRCT的传统影像学征象
Table 1. Basic information and traditional imaging features of 65 patients with lung adenocarcinoma
-
每例患者共提取到1308个影像组学特征,通过单因素方差分析对组学数据进行初次降维,得到微血管浸润阳性与阴性患者间差异有统计学意义的影像组学特征共142个,通过Lasso-Logistic回归分析进一步降维,最终得到6个最优影像组学特征。影像组学得分计算公式见公式(1)。绘制的影像组学得分图见图2。
图 2 65例肺腺癌患者的HRCT影像组学得分图
Figure 2. Radiomic scores histogram of 65 patients with lung adenocarcinoma
$\begin{split} {\rm{Radscore}}=&0.077 \times {\rm{square.glszm.SizeZoneNonUniformityNormalized}}- \\ &0.111 \times {\rm{squareroot.firstorder.MeanAbsoluteDeviation}}-\\ &0.230 \times {\rm{squareroot.glcm.SumSquares}}+ \\ &0.020\times {\rm{wavelet\_HHH.glszm.LargeAreaEmphasis}}+ \\ &0.173 \times{\rm{wavelet\_HLH.gldm.LargeDependenceHighGrayLevelEmphasis}}+\\ &0.199 \times {\rm{wavelet\_HLH.glszm.SizeZoneNonUniformityNormalized}}\end{split}$ -
联合训练组中6个最优影像组学特征和传统影像学征象得到综合模型,绘制的列线图见图3。通过ROC曲线(图4)评估综合模型预测肺腺癌微血管浸润的效能,结果见表2。综合模型在训练组中的AUC为0.880(95%CI:0.750~0.957)、灵敏度为90.5%、特异度为72.0%;在验证组中的AUC为0.811(95%CI:0.568~0.951)、灵敏度为88.9%、特异度为80.0%。与传统影像学征象及影像组学得分相比,综合模型具有更好的诊断效能。
项目 训练组(n=46) 验证组(n=19) AUC(95%CI) 灵敏度(%) 特异度(%) AUC(95%CI) 灵敏度(%) 特异度(%) 肿瘤最大径 0.648(0.493~0.783) 38.1 88.0 0.783(0.538~0.936) 88.9 70.0 影像组学得分 0.870(0.739~0.951) 90.5 72.0 0.800(0.556~0.945) 88.9 90.0 综合模型a 0.880(0.750~0.957) 90.5 72.0 0.811(0.568~0.951) 88.9 80.0 注:表中,a:基于影像组学联合传统影像学征象的综合模型;AUC:曲线下面积;CI:可变区间 表 2 肺腺癌微血管浸润预测模型的受试者工作特征曲线分析结果
Table 2. Receiver operating characteristic curve analysis results of microvascular invasion prediction model for lung adenocarcinoma
高分辨率CT影像组学联合传统影像学征象预测肺腺癌微血管浸润的价值
Value of HRCT radiomics combined with traditional imaging features in predicting microvascular invasion of lung adenocarcinoma
-
摘要:
目的 探讨高分辨率CT(HRCT)影像组学联合传统影像学征象的综合模型预测肺腺癌微血管浸润的价值。 方法 回顾性分析2015年6月至2019年4月于青岛大学附属医院就诊的微血管浸润状态明确的肺腺癌患者65例(微血管浸润阳性30例、阴性35例),其中,男性33例、女性32例,年龄34~83(60.7±10.3)岁。以患者HRCT检查时间为编号,通过系统随机抽样方法将患者按约3∶1等距抽样分为2组:训练组46例,验证组19例。训练组用于模型的建立,验证组用于模型的效能评价。通过两独立样本t检验、χ2检验或Fisher确切概率法筛选训练组中微血管浸润阳性与阴性患者间差异有统计学意义的传统影像学征象。勾画2组患者的肿瘤三维感兴趣区并提取影像组学特征,通过单因素方差分析和Lasso-Logistic回归分析筛选训练组中有鉴别价值的最优影像组学特征,计算影像组学得分。通过Logistic回归分析构建联合影像组学得分和传统影像学征象预测肺腺癌微血管浸润的综合模型,并绘制列线图,进行效能评价。 结果 共提取影像组学特征1308个,最终得到6个最优影像组学特征。传统影像学征象中仅肿瘤最大径在微血管浸润阳性与阴性患者间的差异有统计学意义[(28.10±11.39)mm对(22.32±6.26) mm;t=5.580,P=0.035],其在训练组中的曲线下面积(AUC)为0.648(95%CI:0.493~0.783)、灵敏度为38.1%、特异度为88.0%;在验证组中的AUC为0.783(95%CI:0.538~0.936)、灵敏度为88.9%、特异度为70.0%。预测肺腺癌微血管浸润的综合模型在训练组中的AUC为0.880(95%CI:0.750~0.957),灵敏度为90.5%,特异度为72.0%;在验证组中的AUC为0.811(95%CI:0.568~0.951),灵敏度为88.9%,特异度为80.0%。 结论 基于HRCT影像组学联合传统影像学征象的综合模型对肺腺癌微血管浸润具有较高的预测价值,有助于肺腺癌患者的术前评估。 -
关键词:
- 肺腺癌 /
- 体层摄影术,X线计算机 /
- 影像组学 /
- 微血管浸润
Abstract:Objective To explore the value of the radiomic nomogram integrating high-resolution CT (HRCT) radiomics and traditional imaging features to predict the microvascular invasion (MVI) of lung adenocarcinoma (LAC). Methods A total of 65 patients with LAC (30 MVI-present LACs and 35 MVI-absent LACs) with pathologically confirmed MVI status in the Affiliated Hospital of Qingdao University from June 2015 to April 2019 were retrospectively enrolled, among whom 33 were males and 32 were females with age range of 34–83 (60.7±10.3) years old. After patients were numbered depending on their HRCT examination time, they were randomly divided into two groups according to the systematic sampling method (about 3∶1 ratio equidistance sampling): 46 patients constituted the training set and 19 patients constituted the validation set. The training set was used to build the model, and the validation set was used to evaluate the model effectiveness. The traditional imaging features with a significant difference between the MVI-present and MVI-absent patients were selected by two independent samples t test, χ2 test, or Fisher's exact probability method. The three-dimensional regions of interest of the tumors in the two groups were drawn, and the radiomic features were extracted. The optimal radiomic features were selected by one-way ANOVA and Lasso-Logistic regression analysis, and the radiomic scores were calculated. The combined nomogram to predict MVI of LAC, incorporating the radiomic scores and the traditional imaging features, was constructed by Logistic regression, and its effectiveness was evaluated. Results A total of 1308 radiomic features were extracted, and 6 optimal radiomic features were finally obtained. Among the traditional imaging features, only the longest diameter of the tumor was statistically different between the MVI-present and the MVI-absent patients [(28.10±11.39) mm vs. (22.32±6.26) mm; t=5.580, P=0.035]. For the traditional imaging features, the area under the curve (AUC) was 0.648 (95%CI: 0.493–0.783), the sensitivity was 38.1%, and the specificity was 88.0% in the training set; meanwhile, the AUC was 0.783 (95%CI: 0.538–0.936), the sensitivity was 88.9%, and the specificity was 70.0% in the validation set. For the combined nomogram, the AUC was 0.880 (95%CI: 0.750–0.957), the sensitivity was 90.5%, and the specificity was 72.0% in the training set; whereas the AUC was 0.811 (95%CI: 0.568–0.951), the sensitivity was 88.9%, and the specificity was 80.0% in the validation set. Conclusion The radiomic nomogram, incorporating HRCT radiomics and traditional imaging features, shows favorable predictive efficacy for MVI status in LAC, which might assist in the preoperative evaluation of patients with LAC. -
Key words:
- Adenocarcinoma of lung /
- Tomography, X-ray computed /
- Radiomics /
- Microvascular invasion
-
表 1 65例肺腺癌患者的临床资料及HRCT的传统影像学征象
Table 1. Basic information and traditional imaging features of 65 patients with lung adenocarcinoma
项目 训练组(n=46) 验证组(n=19) 微血管浸润
阳性(n=21)微血管浸润
阴性(n=25)检验值 P值 微血管浸润
阳性(n=9)微血管浸润
阴性(n=10)检验值 P值 男性/女性(例) 11/10 8/17 χ2=1.955 0.162 5/4 9/1 χ2=2.898 0.238 年龄(岁) 62.90±12.57 59.36±8.09 t=4.574 0.254 60.78±12.97 59.20±7.93 t=4.169 0.750 肿瘤最大径(mm) 28.10±11.39 22.32±6.26 t=5.580 0.035 32.56±11.05 23.40±9.87 t=0.128 0.073 所在肺(左肺/右肺,例) 9/12 10/15 χ2=0.038 0.845 2/7 1/9 χ2=0.532 0.921 所在肺叶(上/中/下,例) 14/1/6 18/2/5 χ2=0.581 0.748 6/1/2 6/1/3 χ2=0.148 0.929 形状(类圆/不规则,例) 13/8 15/10 χ2=0.017 0.895 5/4 8/2 χ2=1.310 0.515 成分(实性/磨玻璃及混杂,例) 18/3 17/8 χ2=1.968 0.161 9/0 9/1 χ2=0.950 1.000 分叶征(有/无,例) 20/1 25/0 χ2=1.217 0.457 9/0 9/1 χ2=0.950 1.000 毛刺征(有/无,例) 10/11 17/8 χ2=1.955 0.162 6/3 8/2 χ2=0.434 0.891 空洞(有/无,例) 4/17 4/21 χ2=0.074 1.000 1/8 1/9 χ2=0.006 1.000 支气管充气征(有/无,例) 4/17 9/16 χ2=1.618 0.203 2/7 5/5 χ2=1.571 0.437 胸膜凹陷征(有/无,例) 15/6 20/5 χ2=0.461 0.497 7/2 8/2 χ2=0.014 1.000 血管集束征(有/无,例) 12/9 19/6 χ2=1.847 0.174 8/1 10/0 χ2=1.173 0.474 注:表中,HRCT:高分辨率CT 表 2 肺腺癌微血管浸润预测模型的受试者工作特征曲线分析结果
Table 2. Receiver operating characteristic curve analysis results of microvascular invasion prediction model for lung adenocarcinoma
项目 训练组(n=46) 验证组(n=19) AUC(95%CI) 灵敏度(%) 特异度(%) AUC(95%CI) 灵敏度(%) 特异度(%) 肿瘤最大径 0.648(0.493~0.783) 38.1 88.0 0.783(0.538~0.936) 88.9 70.0 影像组学得分 0.870(0.739~0.951) 90.5 72.0 0.800(0.556~0.945) 88.9 90.0 综合模型a 0.880(0.750~0.957) 90.5 72.0 0.811(0.568~0.951) 88.9 80.0 注:表中,a:基于影像组学联合传统影像学征象的综合模型;AUC:曲线下面积;CI:可变区间 -
[1] Ettinger DS, Akerley W, Borghaei H, et al. Non-small cell lung cancer, version 2.2013[J]. J Natl Compr Canc Netw, 2013, 11(6): 645−653. DOI: 10.6004/jnccn.2013.0084. [2] Usui S, Minami Y, Shiozawa T, et al. Differences in the prognostic implications of vascular invasion between lung adenocarcinoma and squamous cell carcinoma[J]. Lung Cancer, 2013, 82(3): 407−412. DOI: 10.1016/j.lungcan.2013.09.001. [3] 张静渊, 尹荣, 徐新宇. 非小细胞肺癌的脉管内侵犯与微转移、微血管密度关系的研究[J]. 南京医科大学学报: 自然科学版, 2013, 33(7): 887−891. DOI: 10.7655/NYDXBNS20130706.
Zhang JY, Yin R, Xu XY. Microvessel density and micrometastasis in patients with vascular-invaded non-small cell lung cancer[J]. J Nanjing Med Univ: Natur Sci, 2013, 33(7): 887−891. DOI: 10.7655/NYDXBNS20130706.[4] Bodendorf MO, Haas V, Laberke HG, et al. Prognostic value and therapeutic consequences of vascular invasion in non-small cell lung carcinoma[J]. Lung Cancer, 2009, 64(1): 71−78. DOI: 10.1016/j.lungcan.2008.07.011. [5] Patel AJ, Daniel G, Naidu B, et al. The significance of microvascular invasion after complete resection of early-stage non-small-cell lung cancer[J]. Interact Cardiovasc Thorac Surg, 2016, 22(1): 101−105. DOI: 10.1093/icvts/ivv287. [6] Gillies RJ, Kinahan PE, Hricak H. Radiomics: images are more than pictures, they are data[J]. Radiology, 2016, 278(2): 563−577. DOI: 10.1148/radiol.2015151169. [7] Balagurunathan Y, Gu YH, Wang H, et al. Reproducibility and prognosis of quantitative features extracted from CT images[J/OL]. Transl Oncol, 2014, 7(1): 72−87[2019-09-02]. https://www.sciencedirect.com/science/article/pii/S1936523314800107. DOI: 10.1593/tlo.13844. [8] Kumar V, Gu YH, Basu S, et al. Radiomics: the process and the challenges[J]. Magn Reson Imaging, 2012, 30(9): 1234−1248. DOI: 10.1016/j.mri.2012.06.010. [9] Wu Y, Xu L, Yang PF, et al. Survival prediction in high-grade osteosarcoma using radiomics of diagnostic computed tomography[J/OL]. EBioMedicine, 2018, 34: 27−34[2019-09-02]. http://www.ncbi.nlm.nih.gov/pmc/articles/PMC6116348. DOI: 10.1016/j.ebiom.2018.07.006. [10] 温焕舜, 梁朝阳. 非小细胞肺癌肿瘤组织血管浸润对预后的影响[J/OL]. 中华胸部外科电子杂志, 2016, 3(4): 239−242[2019-09-02]. http://rs.yiigle.com/CN119357201604/1072582.htm. DOI: 10.3877/cma.j.issn.2095-8773.2016.04.010.
Wen HS, Liang CY. Impact of blood vessel invasion on prognosis of non-small cell lung cancer[J/OL]. Chin J Thorac Surg (Electron Ed), 2016, 3(4): 239−242[2019-09-02]. http://rs.yiigle.com/CN119357201604/1072582.htm. DOI: 10.3877/cma.j.issn.2095-8773.2016.04.010.[11] Tsubokawa N, Mimae T, Miyata Y, et al. Prognostic significance of vascular invasion in intermediate-grade subtype of lung adenocarcinoma[J]. Jpn J Clin Oncol, 2016, 46(11): 1015−1021. DOI: 10.1093/jjco/hyw113. [12] Hishida T, Yoshida J, Maeda R, et al. Prognostic impact of intratumoural microvascular invasion and microlymphatic permeation on node-negative non-small-cell lung cancer: which indicator is the stronger prognostic factor?[J]. Eur J Cardiothorac Surg, 2013, 43(4): 772−777. DOI: 10.1093/ejcts/ezs396. [13] Wang J, Wang BC, Bi JW, et al. Prognostic significance of microvascular invasion and microlymphatic permeation in non-small-cell lung cancer[J]. Eur J Cardiothorac Surg, 2013, 43(6): 1269. DOI: 10.1093/ejcts/ezs632. [14] Hamanaka R, Yokose T, Sakuma Y, et al. Prognostic impact of vascular invasion and standardization of its evaluation in stage I non-small cell lung cancer[J/OL]. Diagn Pathol, 2015, 10: 17[2019-09-02]. https://diagnosticpathology.biomedcentral.com/articles/10.1186/s1300-015-0249-5. DOI: 10.1186/s13000-015-0249-5. [15] National Comprehensive Cancer Network. Non-Small Cell Lung Cancer, NCCN Guidelines Version 2.2019[EB/OL]. Fort Washington: NCCN, 2018 (2018-11-21)[2019-09-02]. http://www.nccn.org/patients/guidelines/cancers.aspx#lung. [16] Owens CA, Peterson CB, Tang C, et al. Lung tumor segmentation methods: impact on the uncertainty of radiomics features for non-small cell lung cancer[J/OL]. PLoS One, 2018, 13(10): e0205003[2019-09-02]. https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0205003. DOI: 10.1371/journal.pone.0205003. [17] Shi Z, Deng JJ, She YL, et al. Quantitative features can predict further growth of persistent pure ground-glass nodule[J]. Quant Imaging Med Surg, 2019, 9(2): 283−291. DOI: 10.21037/qims.2019.01.04. [18] Wu WM, Parmar C, Grossmann P, et al. Exploratory study to identify radiomics classifiers for lung cancer histology[J/OL]. Front Oncol, 2016, 6: 71[2019-09-02]. http://www.frontiersin.org/articles/10.3389/fonc.2016.00071/full. DOI: 10.3389/fonc.2016.00071.