-
肺癌是世界上癌症相关死亡的主要原因,其发病率和死亡率均居全球第一[1]。其中非小细胞肺癌(non-small cell lung cancer, NSCLC)占肺癌的80%~85%[2]。国家综合癌症指南(2018版)指出,NSCLC的淋巴结转移状态,对术式的选择、淋巴结清扫范围、是否需要新辅助化疗及疾病预后至关重要[3]。目前胸腔镜检查或超声引导下穿刺活检是评价NSCLC淋巴结转移的主要方法,结果较准确,但有创。无创且能有效评估NSCLC病人淋巴结状态的检查方法越来越受到临床的关注。近年来,影像组学在医学领域发展迅速,它可以反映关于肿瘤的生物学信息,并可以无创地提供关于诊断、预后评估和治疗反应预测的信息[4]。研究[5-8]表明,对于术前预测淋巴结转移,影像组学已经应用于头颈部癌、结直肠癌、胃癌等全身多个系统,然而,对预测肺癌淋巴结转移的报道较少。本研究旨在探讨CT影像组学在预测NSCLC病人淋巴结转移中的价值,并与传统临床危险因素和影像主观征象进行对比。
-
训练组和验证组有淋巴结转移病人的病灶最长径大于无淋巴结转移病人(P < 0.05),毛刺征发生比例高于无转移病人(P < 0.01和P < 0.05),其他特征在有无淋巴结转移病人之间差异均无统计学意义(P>0.05)(见表 1)。
特征 训练组(n = 100) 验证组(n = 43) 无淋巴结转移(n=66) 有淋巴结转移(n=34) χ2 P 无淋巴结转移(n=29) 有淋巴结转移(n=14) χ2 P 年龄(x±s)/岁 64.36±9.18 62.15±9.90 1.11* >0.05 65.28±7.82 64.57±8.04 0.27* >0.05 性别 男
女48(72.73)
18(27.27)24(70.59)
10(29.41)0.05 >0.05 20(68.97)
9(31.03)10(71.43)
4(28.57)0.00 >0.05 基础病 无
有38(57.58)
28(42.42)26(76.47)
8(23.53)3.48 >0.05 18(62.07)
11(37.93)8(57.14)
6(42.86)0.96 >0.05 吸烟史 无
有43(65.15)
23(34.85)17(50.00)
17(50.00)2.15 >0.05 21(72.41)
8(27.59)9(64.29)
5(35.71)0.30 >0.05 原发肿瘤病史 无
有60(90.91)
6(9.09)32(94.12)
2(5.88)0.03 >0.05 29(100.00)
0(0.00)12(85.71)
2(14.29)1.72 >0.05 病灶形态 圆形/类圆形
不规则3(4.55)
63(95.45)0(0.00)
34(100.00)0.41 >0.05 1(3.45)
28(96.55)0(0.00)
14(100.00)0.00 >0.05 分叶征 无
有2(3.03)
64(96.97)4(11.76)
30(88.24)1.68 >0.05 1(3.45)
28(96.55)0(0.00)
14(100.00)0.00 >0.05 毛刺征 无
有43(65.15)
23(34.85)13(38.23)
21(61.77)6.60 < 0.01 19(79.17)
10(20.83)4(21.05)
10(80.95)5.18 < 0.05 血管集束征 无
有61(92.42)
5(7.58)33(97.06)
1(2.94)0.23 >0.05 25(86.21)
4(13.79)13(92.86)
1(7.14)0.17 >0.05 胸膜凹陷征 无
有43(65.15)
23(34.85)26(76.47)
8(23.53)1.34 >0.05 18(62.07)
11(37.93)9(64.29)
5(35.71)0.20 >0.05 空洞/空泡征 无
有59(89.39)
7(10.61)29(85.29)
5(14.71)0.36 >0.05 23(79.31)
6(20.69)13(92.86)
1(7.14)0.47 >0.05 液化/坏死 无
有55(83.33)
11(16.67)29(85.29)
5(14.71)0.06 >0.05 26(89.66)
3(10.34)13(92.86)
1(7.14)0.00 >0.05 钙化 无
有60(90.91)
6(9.09)33(97.06)
1(2.94)0.53 >0.05 26(89.66)
3(10.34)14(100.00)
0(0.00)0.37 >0.05 病灶位置 上叶
中下叶30(45.45)
36(54.55)10(29.41)
24(70.59)2.41 >0.05 15(51.72)
14(48.28)11(78.57)
3(21.43)1.84 >0.05 *示t值 表 1 病人临床因素及CT征象统计学比较[n; 构成比(%)]
-
临床模型由病灶最长径和毛刺征组成,其回归方程为:P=1.074×毛刺征+0.442×最长径-2.501。通过比较预测值P(Predict)与约登指数的大小来判断淋巴结是否转移,当P值大于约登指数时,认为有淋巴结转移,反之无淋巴结转移,该模型的约登指数为0.26。
影像组学预测模型由6个最优影像组学特征组成,分别为:(1)灰度共生矩阵(GLCM)特征4个:基于平方根滤波的逆方差(squareroot-glcm-inverse variance, Squareroot-IV)、基于平方滤波的逆方差(square-glcm-inverse variance, Square-IV)、基于指数滤波的最大相关系数(exponential-MCC, E-MCC)、基于平方根滤波的最大相关系数(squareroot-MCC, S-MCC);(2)小波特征:基于小波变换的峰度(firstorder-kurtosis, FK);(3)灰度依赖矩阵:基于指数滤波的低灰度强化小依赖度(small dependence low gray level emphasis, SDLGLE)。放射组学评分,其公式为:RadScore =-1.578×(Squareroot-IV)-1.510×FK-1.228×(E-MCC)-1.145×(Square - IV)+1.069×SDLGLE-0.911×(S-MCC)+2.606。通过计算机调整惩罚系数后,得到最佳系数对应的放射组学评分界值为-7.475,当放射组学评分大于-7.475时,判定有淋巴结转移,反之,无淋巴结转移。
-
影像组学模型的ROC曲线下面积(AUC)高于临床模型(P < 0.01和P < 0.05)(见表 2、图 4)。
模型 训练组 验证组 AUC(95%CI) 准确度/% 敏感性/% 特异性/% AUC(95%CI) 准确度/% 敏感性/% 特异性/% 临床模型 0.662(0.561~0.754) 62.0 32.4 93.9 0.664(0.504~0.800) 60.0 100.0 44.8 影像组学模型 0.864(0.781~0.924) 79.0 85.3 72.7 0.860(0.720~0.946) 81.0 85.7 82.8 Z 4.36 2.03 P < 0.01 < 0.05 表 2 2种模型在训练组和验证组中的AUC、准确度、敏感性、特异性的比较
CT影像组学在预测非小细胞肺癌淋巴结转移中的价值
The value of CT radiomics in the prediction of lymph node metastasis in non-small cell lung cancer
-
摘要:
目的探讨基于胸部CT影像组学在术前预测非小细胞肺癌淋巴结转移中的价值。 方法回顾性分析经术后病理证实的143例非小细胞肺癌病人临床、胸部CT增强影像资料。按照7:3比例,随机分为训练组(n=100)和验证组(n=43)。在静脉期图像上提取肿瘤的影像组学特征,采用最小绝对收缩选择算子(LASSO)逻辑回归用于数据降维、特征筛选。分别基于影像组学特征和临床-影像特征(最大径、毛刺征)构建预测模型。采用受试者工作特征曲线(ROC)的曲线下面积(AUC)评价不同模型的鉴别预测效能,并对模型的ROC曲线行Delong检验;在验证组中评估其预测效能。 结果共提取939个影像组学特征,经筛选最终得到6个最优特征并建立影像组学预测模型。对于术前预测淋巴结转移的效能,在训练组中,影像组学模型AUC为0.864(95%CI:0.781~0.924),大于临床模型的AUC为0.662(95%CI:0.561~0.754)(P < 0.01);在验证组中,影像组学模型AUC为0.860(95%CI:0.720~0.964),大于临床模型的AUC为0.664(95%CI:0.504~0.880)(P < 0.05)。 结论基于胸部CT增强图像提取影像组学特征及其构建的预测模型,影像组学模型的效能高于临床模型,可以作为一种预测非小细胞癌病人淋巴结是否转移的辅助工具,具有良好的临床应用前景。 Abstract:ObjectiveTo explore the value of chest enhanced CT radiomics in the prediction of lymph node metastasis in patients with non-small cell lung cancer(NSCLC). MethodsThe clinical and chest enhanced CT data of 143 NSCLC patients confirmed by pathologically were retrospectively analyzed.The patients were randomly divided into the training group(n=100) and verification group(n=43) according to the ratio of 7:3.The venous phase images were used to extract the radiomics features.The least absolute shrinkage and selection operator(LASSO) logistic regression was used for data dimension reduction and feature selection.Two predictive models were constructed using the radiomics features and clinical-imaging characteristics(the maximum meridian and burr sign).The AUCs of ROC was used to evaluate the predictive effectiveness of model.The ROC curve of model was tested by Delong test.The predictive efficacy was evaluated in validation group. ResultsA total of 939 radiomics features were extracted, 6 optimal features were finally selected, and the prediction model was established.In the training group, the AUC of the radiomics model was 0.864(95%CI: 0.781~0.924), which was higher than that of the clinical model[0.662 (95%CI: 0.561~0.754)](P < 0.01).In the validation group, the AUC of the radiomics model was 0.860(95%CI: 0.720~0.964), which was greater than that of the clinical model[(0.664 (95%CI: 0.504~0.880)](P < 0.05). ConclusionsBased on the image omics features extracted from the chest CT enhanced images and constructing the prediction model, the efficacy of the radiomics model was higher than that of the clinical model.The CT radiomics can be used as an auxiliary tool to predict lymph node metastasis in patients with non-small cell cancer, which has a good clinical application prospect. -
Key words:
- radiomics /
- non-small cell lung cancer /
- lymph node metastasis
-
表 1 病人临床因素及CT征象统计学比较[n; 构成比(%)]
特征 训练组(n = 100) 验证组(n = 43) 无淋巴结转移(n=66) 有淋巴结转移(n=34) χ2 P 无淋巴结转移(n=29) 有淋巴结转移(n=14) χ2 P 年龄(x±s)/岁 64.36±9.18 62.15±9.90 1.11* >0.05 65.28±7.82 64.57±8.04 0.27* >0.05 性别 男
女48(72.73)
18(27.27)24(70.59)
10(29.41)0.05 >0.05 20(68.97)
9(31.03)10(71.43)
4(28.57)0.00 >0.05 基础病 无
有38(57.58)
28(42.42)26(76.47)
8(23.53)3.48 >0.05 18(62.07)
11(37.93)8(57.14)
6(42.86)0.96 >0.05 吸烟史 无
有43(65.15)
23(34.85)17(50.00)
17(50.00)2.15 >0.05 21(72.41)
8(27.59)9(64.29)
5(35.71)0.30 >0.05 原发肿瘤病史 无
有60(90.91)
6(9.09)32(94.12)
2(5.88)0.03 >0.05 29(100.00)
0(0.00)12(85.71)
2(14.29)1.72 >0.05 病灶形态 圆形/类圆形
不规则3(4.55)
63(95.45)0(0.00)
34(100.00)0.41 >0.05 1(3.45)
28(96.55)0(0.00)
14(100.00)0.00 >0.05 分叶征 无
有2(3.03)
64(96.97)4(11.76)
30(88.24)1.68 >0.05 1(3.45)
28(96.55)0(0.00)
14(100.00)0.00 >0.05 毛刺征 无
有43(65.15)
23(34.85)13(38.23)
21(61.77)6.60 < 0.01 19(79.17)
10(20.83)4(21.05)
10(80.95)5.18 < 0.05 血管集束征 无
有61(92.42)
5(7.58)33(97.06)
1(2.94)0.23 >0.05 25(86.21)
4(13.79)13(92.86)
1(7.14)0.17 >0.05 胸膜凹陷征 无
有43(65.15)
23(34.85)26(76.47)
8(23.53)1.34 >0.05 18(62.07)
11(37.93)9(64.29)
5(35.71)0.20 >0.05 空洞/空泡征 无
有59(89.39)
7(10.61)29(85.29)
5(14.71)0.36 >0.05 23(79.31)
6(20.69)13(92.86)
1(7.14)0.47 >0.05 液化/坏死 无
有55(83.33)
11(16.67)29(85.29)
5(14.71)0.06 >0.05 26(89.66)
3(10.34)13(92.86)
1(7.14)0.00 >0.05 钙化 无
有60(90.91)
6(9.09)33(97.06)
1(2.94)0.53 >0.05 26(89.66)
3(10.34)14(100.00)
0(0.00)0.37 >0.05 病灶位置 上叶
中下叶30(45.45)
36(54.55)10(29.41)
24(70.59)2.41 >0.05 15(51.72)
14(48.28)11(78.57)
3(21.43)1.84 >0.05 *示t值 表 2 2种模型在训练组和验证组中的AUC、准确度、敏感性、特异性的比较
模型 训练组 验证组 AUC(95%CI) 准确度/% 敏感性/% 特异性/% AUC(95%CI) 准确度/% 敏感性/% 特异性/% 临床模型 0.662(0.561~0.754) 62.0 32.4 93.9 0.664(0.504~0.800) 60.0 100.0 44.8 影像组学模型 0.864(0.781~0.924) 79.0 85.3 72.7 0.860(0.720~0.946) 81.0 85.7 82.8 Z 4.36 2.03 P < 0.01 < 0.05 -
[1] XIE Y, ZHAO H, GUO Y, et al. A PET/CT nomogram incorporating SUVmax and CT radiomics for preoperative nodal staging in non-small cell lung cancer[J]. Eur Radiol, 2021, 31(8): 6030. doi: 10.1007/s00330-020-07624-9 [2] SIEGEL RL, MILLER KD, JEMAL A. Cancer statistics, 2016[J]. CA Cancer J Clin, 2016, 66(1): 7. doi: 10.3322/caac.21332 [3] ETTINGER DS, WOOD DE, AISNER DL, et al. Non-small cell lung cancer, version 5.2017, NCCN clinical practice guidelines in oncology[J]. J Natl Compr Canc Netw, 2017, 15(4): 504. doi: 10.6004/jnccn.2017.0050 [4] HUANG Y, LIU Z, HE L, et al. Radiomics signature: a potential biomarker for the prediction of disease-free survival in early-stage (Ⅰ or Ⅱ) non-small cell lung cancer[J]. Radiology, 2016, 281(3): 947. doi: 10.1148/radiol.2016152234 [5] LIU T, ZHOU S, YU J, et al. Prediction of lymph node metastasis in patients with papillary thyroid carcinoma: a radiomics method based on preoperative ultrasound images[J]. Technol Cancer Res Treat, 2019, 18: 1. [6] 贾传亮, 曹嫒, 宋晴, 等. 基于MR的影像组学列线图预测喉癌颈淋巴结转移的临床价值[J]. 中华耳鼻咽喉头颈外科杂志, 2020, 55(12): 1154. doi: 10.3760/cma.j.cn115330-20200719-00604 [7] LI M, ZHANG J, DAN Y, et al. A clinical-radiomics nomogram for the preoperative prediction of lymph node metastasis in colorectal cancer[J]. J Transl Med, 2020, 18(1): 46. doi: 10.1186/s12967-020-02215-0 [8] 王小雷, 高玉青, 徐鹤, 等. 基于能谱CT纹理分析在预测胃癌术前淋巴结转移中的价值[J]. 蚌埠医学院学报, 2021, 46(1): 21. [9] THAWANI R, MCLANCE M, BEIG N, et al. Radiomics and radiogenomics in lung cancer: a review for the clinician[J]. Lung Cancer, 2018, 115: 34. doi: 10.1016/j.lungcan.2017.10.015 [10] CONG M, YAO H, LIU H, et al. Development and evaluation of a venous computed tomography radiomics model to predict lymph node metastasis from non-small cell lung cancer[J]. Medicine (Baltimore), 2020, 99(18): e20074. doi: 10.1097/MD.0000000000020074 [11] YANG G, NIE P, ZHAO L, et al. 2D and 3D texture analysis to predict lymph ovascular invasion in lung adenocarcinoma[J]. Eur J Radiol, 2020, 129: 109111. doi: 10.1016/j.ejrad.2020.109111 [12] DENG HY, ZENG M, LI G, et al. Lung adenocarcinoma has a higher risk of lymph node metastasis than Squamous cell carcinoma: a propensity score-matched analysis[J]. World J Surg, 2019, 43(3): 955. doi: 10.1007/s00268-018-4848-7 [13] 王敏, 宋彬, 黄子星, 等. 大数据时代的精准影像医学: 放射组学[J]. 中国普外基础与临床杂志, 2016, 23(6): 752. [14] ZHONG Y, YUN M, ZHANG T, et al. Radiomics approach to prediction of occult mediastinal lymph node metastasis of lung adenocarcinoma[J]. AJR Am J Roentgenol, 2018, 211(1): 109. doi: 10.2214/AJR.17.19074 [15] LIU Y, KIM J, BALAGURUNATHAN Y, et al. Prediction of pathological nodal involvement by CT-based radiomics features of the primary tumor in patients with clinically node-negative peripheral lung adenocarcinomas[J]. Med Phys, 2018, 45(6): 2518. doi: 10.1002/mp.12901 [16] PHAN TD, WATANABE Y, HIGUCHI M, et al. Texture analysis and synthesis of malignant and benign mediastinal lymph nodes in patients with lung cancer on computed tomography[J]. Sci Rep, 2017, 1(1): 43209. [17] ANDERSEN MB, HARDERS SW, GANESHAN B, et al. CT texture analysis can help differentiate between malignant and benign lymph nodes in the mediastinum in patients suspected for lung cancer[J]. Acta Radiol, 2016, 57(6): 669. doi: 10.1177/0284185115598808 [18] YANG X, PAN X, LIU H, et al. A new approach to predict lymph node metastasis in solid lung adenocarcinoma: a radiomics nomogram[J]. J Thorac Dis, 2018, 10(Suppl 7): S807.