en
×

分享给微信好友或者朋友圈

使用微信“扫一扫”功能。
通讯作者:

于瑞磊,E⁃mail:yuruilei520184@163.com

中图分类号:R814.42

文献标识码:A

文章编号:1007-4368(2021)04-1063-06

DOI:10.7655/NYDXBNS20210721

参考文献 1
夏宁,张宇,郝可可,等.血清肿瘤标志物联合检测诊断肺癌的临床应用研究[J].南京医科大学学报(自然科学版),2015,35(12):1784-1786
参考文献 2
CHU Z G,SHENG B,LIU M Q,et al.Differential diagno⁃ sis of solitary pulmonary inflammatory lesions and periph⁃ eral lung cancers with contrast ⁃ enhanced computed to⁃ mography[J].Clinics(Sao Paulo),2016,71(10):555-561
参考文献 3
李明英,张成琪,邓凯.CT能谱成像对肺内良恶性肿块诊断的初步研究[J].中华放射学杂志,2013,47(5):410⁃413
参考文献 4
LAMBIN P,RIOS⁃VELAZQUEZ E,LEIJENAAR R,et al.Radiomics:extracting more information from medical im⁃ ages using advanced feature analysis[J].Eur J Cancer,2012,48(4):441-446
参考文献 5
CHEN M L,SHI A H,LI X T,et al.Is there any correla⁃ tion between spectral CT imaging parameters and PD ⁃L1 expression of lung adenocarcinoma?[J].Thorac Cancer,2020,11(2):362-368
参考文献 6
SONG Y,ZHANG J,ZHANG Y D,et al.FeAture explorer(FAE):a tool for developing and comparing radiomics models[J].PLoS One,2020,15(8):e0237587
参考文献 7
ODISIO E G,TRUONG M T,DURAN C,et al.Role of dual ⁃ energy computed tomography in thoracic oncology[J].Radiol Clin North Am,2018,56(4):535-548
参考文献 8
CHAE E J,SONG J W,SEO J B,et al.Clinical utility of dual ⁃ energy CT in the evaluation of solitary pulmonary nodules:initial experience[J].Radiology,2008,249(2):671-681
参考文献 9
WU F,ZHOU H,LI F,et al.Spectral CT imaging of lung cancer:quantitative analysis of spectral parameters and their correlation with tumor characteristics[J].Acad Ra⁃ diol,2018,25(11):1398-1404
参考文献 10
SEIDLER M,FORGHANI B,REINHOLD C,et al.Dual⁃ energy CT texture analysis with machine learning for the evaluation and characterization of cervical lymphadenopa⁃ thy[J].Comput Struct Biotechnol J,2019,17:1009-1015
参考文献 11
ZHANG Y,CHENG J,HUA X,et al.Can spectral CT im⁃ aging improve the differentiation between malignant and benign solitary pulmonary nodules?[J].PLoS One,2016,11(2):e0147537
参考文献 12
HOU W S,WU H W,YIN Y,et al.Differentiation of lung cancers from inflammatory masses with dual⁃energy spec⁃ tral CT imaging[J].Acad Radiol,2015,22(3):337-344
参考文献 13
LIN J Z,ZHANG L,ZHANG C Y,et al.Application of gemstone spectral computed tomography imaging in the characterization of solitary pulmonary nodules:prelimi⁃ nary result[J].J Comput Assist Tomogr,2016,40(6):907-911
参考文献 14
CHOE J,LEE S M,DO K H,et al.Prognostic value of ra⁃ diomic analysis of iodine overlay maps from dual ⁃ energy computed tomography in patients with resectable lung cancer[J].Eur Radiol,2019,29(2):915-923
参考文献 15
BAE J M,JY J,LEE H Y,et al.Pathologic stratification of operable lung adenocarcinoma using radiomics features extracted from dual energy CT images[J].Oncotarget,2017,8(1):523-535
目录contents

    摘要

    目的:探讨并比较能谱CT定量参数及碘基图影像组学特征对肺炎性及恶性病变的鉴别诊断价值。方法:回顾性分析123例(炎性组38例,恶性组85例)行动脉期能谱CT增强扫描的肺结节或肿块患者;比较病灶感兴趣区(region of interest, ROI)在70 keV及40 keV单能量图像下的CT值、能谱曲线斜率(λ)、碘浓度(iodine concentration,IC)、有效原子序数值(effective⁃ Z,Zeff)、标准化碘浓度(normalized iodine concentration,NIC)、标准化有序原子序数值(normalized effective⁃Z,NZeff)及最大径 (D),进行多因素分析并绘制受试者工作特征(receiver operating characteristic,ROC)曲线。手动分割ROI,提取影像组学特征, 以7∶3拆分训练集和测试集,通过皮尔森相关系数法、递归特征消除筛选特征,5折法交叉验证,使用线性判别式分类器建立预测模型,并绘制模型的ROC曲线。结果:炎性组及恶性组在40 keV下的CT值、λ10070 keV、λ7040 keV、λ10040 keV、IC、NIC、Zeff、NZeff以及 D的差异具有统计学意义(P < 0.05),多因素分析提示NIC为独立影响因素(P < 0.001),NIC的曲线下面积(area under curve, AUC)为0.728。影像组学共筛选出10个特征建立预测模型,其AUC为0.843,二者的差异具有统计学意义(P =0.034)。结论: 能谱CT定量参数及碘基图影像组学特征对于肺内炎性及恶性病变的鉴别诊断具有很大价值,碘基图组学预测模型取得更优的诊断效能。

    Abstract

    Objective:To investigate the value of quantitative analysis from spectral CT imaging and radiomics analysis from iodine overlay maps in differentiating inflammatory and malignant pulmonary lesions. Methods:A total of 123 patients with solitary pulmonary nodules underwent contrast enhanced spectral CT scan in arterial phase were divided into two groups(38 cases in inflammatory group and 85 cases in malignant group). The gemstone spectral imaging(GSI)viewer was used for image display and data analysis. The parameters,including CT value on 70 keV and 40 keV monochromatic images,slope of spectrum energy curve(λ),iodine concentration(IC),effective ⁃ Z(Zeff)value,normalized iodine concentration(NIC),normalized effective ⁃ Z(NZeff)value and the maximum diameter of lesion(D)were measured. After the multivariate logistic regression analysis,the individual diagnostic performance of independent factor was compared by receiver operating characteristic(ROC)curve. Radiomics features were extracted from manually segmented ROIs. Patients were randomized devided into training or test set in a ratio of 7∶3. Pearson’s correlation and recursive feature elimination(RFE)were used to select features. A radiomics⁃based predictive model using linear discriminant analysis was developed and calibrated with fivefold cross⁃validation. ROC curve was generated to assess the diagnostic performance. Results: CT value on 40 keV monochromatic image,λ10070 keV,λ7040 keV,λ10040 keV,IC,NIC,Zeff,NZeff and D showed significant differences between inflammatory and malignant lesions(P < 0.05). NIC was an independent factor based on the multivariate logistic regression analysis (P < 0.001). The area under the curve(AUC)value of NIC was 0.728. Ten radiomic features were selected for predicting malignant lesions. The AUC value of model(0.843)was significantly higher than NIC(P =0.034). Conclusion:Spectral CT with GSI mode helps to identify inflammatory and malignant pulmonary lesions. The radiomics ⁃ based predictive model provides a more promising tool for differentiating.

  • 肺结节及肿块的鉴别诊断是临床工作的重点之一,早期诊断及治疗对患者意义重大[1]。传统影像学主要依赖于CT图像上病灶的形态学及增强扫描特征进行诊断[2]。近年来新兴的能谱CT成像技术利用各物质能谱曲线的差异实现碘的物质分离,同时生成单能量图像、碘基图及有效原子序数图,提供更多辅助诊断的定量参数。目前已有学者初步探讨能谱CT成像技术在肺部炎性及恶性病变上的鉴别诊断价值,二者在能谱曲线斜率及碘浓度等定量参数之间存在差异[3]。然而,能谱CT成像的数据信息尚未被充分利用,大量基于影像组学方法的影像组学特征可以为疾病诊疗提供更多信息[4]。本研究在能谱CT定量参数的基础上,探讨基于碘基图的影像组学特征对于肺内炎性及恶性的定性诊断价值,寻求更优的鉴别诊断策略。

  • 1 对象和方法

  • 1.1 对象

  • 回顾性分析2018年8月23日—2019年8月1日于南京医科大学第一附属医院接受能谱CT胸部增强扫描的检查结果,收集表现为结节或肿块影的影像资料。本研究为回顾性研究,经医院伦理委员会批准,患者均签署知情同意书。排除标准:①在扫描前接受新辅助治疗的恶性病变;②非原发性恶性病变;③影像表现呈磨玻璃密度;④图像质量差、伪影重以致影响后续图像分析。最终共纳入123例,男85例,女38例,中位年龄64岁(37~89岁),其中恶性组85例(腺癌52例,鳞癌21例,小细胞癌6例,其他6例),炎性组38例,均经手术、CT引导下细针穿刺活检、经支气管镜活检或后续随访证实。

  • 1.2 方法

  • 1.2.1 扫描方法

  • 扫描设备采用Discovery CT750HD(GE Health⁃ care公司,美国),患者取仰卧位,足先进,于深吸气末屏气时进行扫描,扫描范围自两肺尖至肺底部,经肘静脉团注非离子型对比剂碘普罗胺(300mgI/mL) 90mL,注射流率3mL/s,分别在动脉期(30s)、静脉期(60s)进行扫描,层厚及层距为5.0mm,转速为0.5s/r,矩阵512x512,其中动脉期采用能谱扫描模式(gemstone spectral imaging,GSI),管电压为80~140kVp瞬时切换,管电流自适应,静脉期采用常规扫描方式,管电压为120kVp,管电流为240mA。扫描完成后重建出碘基图、70keV及40keV的单能量图像,重建层厚为1.25mm,并将重建图像及原始数据传入GE AW4.6工作站及影像归档和通信系统 (picture archiving and communication system,PACS)。

  • 1.2.2 能谱CT参数分析

  • 使用GSI viewer浏览器进行分析,采用最大层面全病灶区域法放置感兴趣区(region of interest, ROI),尽可能包括整个病灶,同时避开空洞、坏死组织、支气管、大血管及明显的钙化,分别测量ROI于100、70及40keV单能量图像下的CT值、碘浓度(io⁃ dine concentration,IC)及有效原子序数值(effective⁃ Z,Zeff),同时测量与ROI同层面的降主动脉或左锁骨下动脉的碘浓度(IC0)及有效原子序数值(Zeff0),保证同一层面各ROI的位置及大小完全一致[5]。重复测量3次,结果取平均值。后续的计算包括:① 100-70keV、70-40keV、100-40keV段的能谱曲线斜率λ,公式如下:λ100⁃70keV=|CT值100keV-CT值70keV|/(100-70),λ70⁃40keV=|CT值70keV-CT值40keV|/(70-40),λ100⁃40keV=|CT值 100keV-CT值 40keV|/(100-40)。②标准化碘浓度 (normalized iodine concentration,NIC),公式为NIC=IC/IC0。③标准化有序原子序数值(normalized effec⁃ tive⁃Z,NZeff),公式为NZeff=Zeff/Zeff0。在工作站上测量病灶的轴位最大径(D),精确到小数点后1位小数,重复测量3次取平均值。

  • 1.2.3 影像组学特征提取

  • 以DICOM格式从PACS导出碘基图,在FireVoxel软件中打开,采用最大层面全病灶区域法手动分割ROI,ROI的确定方法与1.2.2所述相同。随后使用内部软件平台提取影像组学特征共107个,包括14个形状特征,18个一阶直方图特征以及75个纹理特征[24个灰度共生矩阵(gray level co⁃occur⁃ rence matrix,GLCM)特征、16个灰度游程矩阵(gray level run⁃length matrix,GLRLM)特征、16个灰度区域大小矩阵(gray level size zone matrix,GLSZM)特征、5个邻域灰度差分矩阵(neighborhood gray⁃tone differ⁃ encematrix,NGTDM)特征、14个灰度相关矩阵(gray level dependence matrix,GLDM)特征]。

  • 1.2.4 数据分析

  • 使用基于Python(3.6.8,https://www.python.org/) 的开源组学分析平台FeAture Explorer(FAE,v0.2.5, https://github.com/salan668/FAE)进行特征分析[6]。由于各特征间数量级差异较大,对数据进行Z⁃score标准化处理,计算公式为Z=(x-x-)/std(x),其中x-为某特征在所有样本中的均值,std(X)为相应的标准差。以7∶3拆分训练集和测试集,训练集86例(良性组/恶性组=27/59),测试集37例(良性组/恶性组=11/26)。由于恶性组病例较多,采用少数类样本合成过采样技术(synthetic minority over⁃sampling tech⁃ nique,SMOTE)使数据均衡化,采样倍率为2.19,采样后的训练集正负样本量达1∶1(良性组/恶性组=59/59)。首先使用皮尔森相关系数法(Pearson cor⁃ relation coefficients,PCC)降维,移除PCC值高于0.86的冗余特征。随后通过递归特征消除(recursive fea⁃ ture elimination,RFE)进一步筛选特征,最终使用线性判别式分类器(linear discriminant analysis,LDA) 建立预测模型,并采用5折交叉验证法验证分类器的性能。

  • 1.3 统计学方法

  • 使用SPSS 21.0软件进行统计学分析,计量资料满足正态性检验则采用 t 检验或校正 t 检验,并以均数±标准差(x- ± s)进行统计描述,否则采用Mann⁃Whitney U 检验,并以中位数及四分位数 [MP25P75)]进行统计描述,多因素分析使用二元Logistic回归,使用MedCalc15.2.2软件绘制受试者特征(receiver operating characteristic,ROC)曲线,评估模型的诊断效能,以曲线下面积(area under curve,AUC)作为评判标准。P <0.05为差异有统计学意义。

  • 2 结果

  • 2.1 能谱CT定量参数分析

  • 恶性组CT值70keV、Zeff、NZeff及两组的D不满足正态性检验(P< 0.05),故采用Mann⁃Whitney U 检验。恶性组与炎性组比较,除CT值70keV外,CT值40keV、 λ100⁃70keV、λ70⁃40keV、λ100⁃40keV、IC、NIC、Zeff、NZeff、D差异均有统计学意义(表1、图1~2)。

  • 表1 两组间能谱CT参数比较

  • Table1 Comparison of spectral CT parameters between the two groups

  • 2.2 预测模型的建立与诊断效能

  • 经PCC降冗后剩余91个特征,经RFE特征选择后最终共筛选出10个影像组学特征用于建立预测模型,包括1个一阶直方图特征(firstorder_Maxi⁃ mum)、2个GLCM特征(glcm_Autocorrelation、gl⁃ cm_JointAverage)、5个GLDM特征(gldm_SmallDe⁃ pendenceHighGrayLevelEmphasis、gldm_LargeDepen⁃ denceLowGrayLevelEmphasis、gldm_LowGrayLevel ⁃ Emphasis、gldm_HighGrayLevelEmphasis、gldm_De⁃ pendenceVariance)、1个GLRLM特征(glrlm_Run⁃ LengthNonUniformity)、1个GLSZM特征 (gl⁃ szm_SmallAreaEmphasis),各特征在模型中的系数详见表2。其中glcm_Autocorrelation、gldm_SmallDe⁃ pendenceHighGr⁃ayLevelEmphasis、glrlm_RunLength⁃ NonUniformity、firstorder_Maximum的特征系数为负值,其余为正值。对预测模型进行ROC曲线分析,以评估其诊断效能(表3)。

  • 2.3 能谱CT定量参数和预测模型的诊断效能比较

  • 将具有统计学意义的能谱CT参数纳入二元Logsitic回归分析,NIC为独立影响因素(P< 0.001)。分别绘制NIC和影像组学预测模型的ROC曲线(图3)并进行比较,预测模型的诊断效能明显优于NIC (Z=2.121,P=0.034,表4)。

  • 图1 腺鳞癌能谱CT图像(恶性组)

  • Fig.1 Spectral CT images of a patient diagnosed pulmonary adenosquamous carcinoma(malignant group)

  • 图2 肉芽肿性炎能谱CT图像(炎性组)

  • Fig.2 Spectral CT images of a patient diagnosed pulmonary granulomatous inflammation(inflammatory group)

  • 表2 特征在预测模型中的系数

  • Table2 The coefficients of features in the model

  • 3 讨论

  • 能谱CT为采用单球管瞬时切换技术的双能量CT成像设备,可以在一次扫描内获取两组完全匹配的高低能量图像,同时生成40~140keV连续的101组单能量图像。由于不同物质存在特定的能谱曲线,根据基物质进行物质分离,可一定程度上实现物质成分的定量评估[7]。碘⁃水分离生成的碘基图可量化碘相关衰减,通过碘浓度反映病变的血供情况,与真实的增强程度相当[8]。目前已有大量研究使用能谱CT图像及碘基图,鉴别良恶性病变、评估肿瘤的生物学行为与预后、评估淋巴结转移情况等[9-11]

  • 本研究使用能谱CT定量参数鉴别肺内炎性及恶性病变。炎性组的CT值40keV、λ100⁃70keV、λ70⁃40keV、λ100⁃40keV、 IC、NIC、Zeff、NZeff均高于恶性组,与既往研究结果一致[312]。70keV的单能量图像可用于模拟管电压120kVp的CT扫描图像,CT值70keV没有统计学差异,提示炎性组与恶性组病变难以直接通过常规增强CT图像的CT值区分。能谱CT可以通过ROI的碘浓度反映血供情况,碘浓度越高,能谱曲线斜率增大、在低能级下的X衰减量增高,从而反映出病变血供之间的差异。病灶动脉期的强化程度主要与血供情况与血流有关。炎性病变主要由增生的肉芽组织构成,毛细血管网丰富,在炎性因子的刺激下,小动脉和毛细血管网明显扩张,血管通透性增高,对比剂可快速进入病灶内,而肿瘤不成熟的血管生成以及癌组织对管壁的浸润会导致血管形态、功能的异常,因此炎性病变较恶性病变呈现更早而明显的均匀强化。本研究使用同层面主动脉或左锁骨下动脉对ROI碘浓度进行标准化,可以最小化患者循环状况以及扫描时间产生的干扰,使用NIC更准确地反映病变的血供情况[13]

  • 表3 ROC曲线分析预测模型在训练集、验证集及测试集下的性能

  • Table3 Performance of the model on training set,validation set and test set in ROC analysis

  • 图3 预测模型的ROC曲线分析

  • Fig.3 ROC curve analysis of the model

  • 表4 ROC曲线分析NIC和预测模型鉴别肺内炎性及恶性病变的性能

  • Table4 Performance of NIC and the model in identifying inflammatory and malignant pulmonary lesions in ROC analysis

  • 在此基础上,本研究基于碘基图提取影像组学特征,建立肺内炎性及恶性病变的预测模型。使用1个一阶直方图特征以及9个纹理特征,以LDA分类器建立模型,其诊断效能较NIC明显提高,ROC曲线存在统计学差异。在预测模型中权重前3的特征分别为glcm_Autocorrelation以及glcm_JointAverage、 gldm_HighGrayLevelEmphasis,其中glcm_Autocorre⁃ lation用于度量纹理的粗细程度,glcm_JointAverage为联合概率的灰度加权和,gldm_HighGrayLevelEm⁃ phasis反映高灰度值的分布情况。碘基图通过显示碘对比剂的分布情况,着重反映病变血供的异质性。既往已有学者尝试使用基于碘基图的影像组学特征判断肺癌预后及病理分级,不同恶性程度的病变之间碘基图的组学特征存在显著差异[14-15]。本文为首次探讨碘基图影像组学特征对于肺炎性及恶性病变鉴别价值的研究,预测模型取得明显优于传统能谱CT定量参数的诊断效能。

  • 本研究探讨并比较能谱CT定量参数以及碘基图影像组学特征对于肺内炎性及恶性病变的预测价值,存在几项局限性:①本研究采用动脉期GSI,未评估静脉期的能谱CT定量参数以及碘基图组学特征的预测价值;②病例数有限,样本量不足,后续有待于大样本的进一步研究;③本研究采用最大层面全病灶区域法,基于最大层面而非全容积进行分析,不能反映病变的全貌,可能会存在一定的偏差,有待今后大样本、基于全容积的研究进一步完善。

  • 参考文献

    • [1] 夏宁,张宇,郝可可,等.血清肿瘤标志物联合检测诊断肺癌的临床应用研究[J].南京医科大学学报(自然科学版),2015,35(12):1784-1786

    • [2] CHU Z G,SHENG B,LIU M Q,et al.Differential diagno⁃ sis of solitary pulmonary inflammatory lesions and periph⁃ eral lung cancers with contrast ⁃ enhanced computed to⁃ mography[J].Clinics(Sao Paulo),2016,71(10):555-561

    • [3] 李明英,张成琪,邓凯.CT能谱成像对肺内良恶性肿块诊断的初步研究[J].中华放射学杂志,2013,47(5):410⁃413

    • [4] LAMBIN P,RIOS⁃VELAZQUEZ E,LEIJENAAR R,et al.Radiomics:extracting more information from medical im⁃ ages using advanced feature analysis[J].Eur J Cancer,2012,48(4):441-446

    • [5] CHEN M L,SHI A H,LI X T,et al.Is there any correla⁃ tion between spectral CT imaging parameters and PD ⁃L1 expression of lung adenocarcinoma?[J].Thorac Cancer,2020,11(2):362-368

    • [6] SONG Y,ZHANG J,ZHANG Y D,et al.FeAture explorer(FAE):a tool for developing and comparing radiomics models[J].PLoS One,2020,15(8):e0237587

    • [7] ODISIO E G,TRUONG M T,DURAN C,et al.Role of dual ⁃ energy computed tomography in thoracic oncology[J].Radiol Clin North Am,2018,56(4):535-548

    • [8] CHAE E J,SONG J W,SEO J B,et al.Clinical utility of dual ⁃ energy CT in the evaluation of solitary pulmonary nodules:initial experience[J].Radiology,2008,249(2):671-681

    • [9] WU F,ZHOU H,LI F,et al.Spectral CT imaging of lung cancer:quantitative analysis of spectral parameters and their correlation with tumor characteristics[J].Acad Ra⁃ diol,2018,25(11):1398-1404

    • [10] SEIDLER M,FORGHANI B,REINHOLD C,et al.Dual⁃ energy CT texture analysis with machine learning for the evaluation and characterization of cervical lymphadenopa⁃ thy[J].Comput Struct Biotechnol J,2019,17:1009-1015

    • [11] ZHANG Y,CHENG J,HUA X,et al.Can spectral CT im⁃ aging improve the differentiation between malignant and benign solitary pulmonary nodules?[J].PLoS One,2016,11(2):e0147537

    • [12] HOU W S,WU H W,YIN Y,et al.Differentiation of lung cancers from inflammatory masses with dual⁃energy spec⁃ tral CT imaging[J].Acad Radiol,2015,22(3):337-344

    • [13] LIN J Z,ZHANG L,ZHANG C Y,et al.Application of gemstone spectral computed tomography imaging in the characterization of solitary pulmonary nodules:prelimi⁃ nary result[J].J Comput Assist Tomogr,2016,40(6):907-911

    • [14] CHOE J,LEE S M,DO K H,et al.Prognostic value of ra⁃ diomic analysis of iodine overlay maps from dual ⁃ energy computed tomography in patients with resectable lung cancer[J].Eur Radiol,2019,29(2):915-923

    • [15] BAE J M,JY J,LEE H Y,et al.Pathologic stratification of operable lung adenocarcinoma using radiomics features extracted from dual energy CT images[J].Oncotarget,2017,8(1):523-535

  • 参考文献

    • [1] 夏宁,张宇,郝可可,等.血清肿瘤标志物联合检测诊断肺癌的临床应用研究[J].南京医科大学学报(自然科学版),2015,35(12):1784-1786

    • [2] CHU Z G,SHENG B,LIU M Q,et al.Differential diagno⁃ sis of solitary pulmonary inflammatory lesions and periph⁃ eral lung cancers with contrast ⁃ enhanced computed to⁃ mography[J].Clinics(Sao Paulo),2016,71(10):555-561

    • [3] 李明英,张成琪,邓凯.CT能谱成像对肺内良恶性肿块诊断的初步研究[J].中华放射学杂志,2013,47(5):410⁃413

    • [4] LAMBIN P,RIOS⁃VELAZQUEZ E,LEIJENAAR R,et al.Radiomics:extracting more information from medical im⁃ ages using advanced feature analysis[J].Eur J Cancer,2012,48(4):441-446

    • [5] CHEN M L,SHI A H,LI X T,et al.Is there any correla⁃ tion between spectral CT imaging parameters and PD ⁃L1 expression of lung adenocarcinoma?[J].Thorac Cancer,2020,11(2):362-368

    • [6] SONG Y,ZHANG J,ZHANG Y D,et al.FeAture explorer(FAE):a tool for developing and comparing radiomics models[J].PLoS One,2020,15(8):e0237587

    • [7] ODISIO E G,TRUONG M T,DURAN C,et al.Role of dual ⁃ energy computed tomography in thoracic oncology[J].Radiol Clin North Am,2018,56(4):535-548

    • [8] CHAE E J,SONG J W,SEO J B,et al.Clinical utility of dual ⁃ energy CT in the evaluation of solitary pulmonary nodules:initial experience[J].Radiology,2008,249(2):671-681

    • [9] WU F,ZHOU H,LI F,et al.Spectral CT imaging of lung cancer:quantitative analysis of spectral parameters and their correlation with tumor characteristics[J].Acad Ra⁃ diol,2018,25(11):1398-1404

    • [10] SEIDLER M,FORGHANI B,REINHOLD C,et al.Dual⁃ energy CT texture analysis with machine learning for the evaluation and characterization of cervical lymphadenopa⁃ thy[J].Comput Struct Biotechnol J,2019,17:1009-1015

    • [11] ZHANG Y,CHENG J,HUA X,et al.Can spectral CT im⁃ aging improve the differentiation between malignant and benign solitary pulmonary nodules?[J].PLoS One,2016,11(2):e0147537

    • [12] HOU W S,WU H W,YIN Y,et al.Differentiation of lung cancers from inflammatory masses with dual⁃energy spec⁃ tral CT imaging[J].Acad Radiol,2015,22(3):337-344

    • [13] LIN J Z,ZHANG L,ZHANG C Y,et al.Application of gemstone spectral computed tomography imaging in the characterization of solitary pulmonary nodules:prelimi⁃ nary result[J].J Comput Assist Tomogr,2016,40(6):907-911

    • [14] CHOE J,LEE S M,DO K H,et al.Prognostic value of ra⁃ diomic analysis of iodine overlay maps from dual ⁃ energy computed tomography in patients with resectable lung cancer[J].Eur Radiol,2019,29(2):915-923

    • [15] BAE J M,JY J,LEE H Y,et al.Pathologic stratification of operable lung adenocarcinoma using radiomics features extracted from dual energy CT images[J].Oncotarget,2017,8(1):523-535