《传染病学传染病学 (4).pdf》由会员分享，可在线阅读，更多相关《传染病学传染病学 (4).pdf（8页珍藏版）》请在taowenge.com淘文阁网|工程机械CAD图纸|机械工程制图|CAD装配图下载|SolidWorks_CaTia_CAD_UG_PROE_设计图分享下载上搜索。
1、Development and Validation of aRadiomic-Based Model forPrediction of IntrahepaticCholangiocarcinoma in Patients WithIntrahepatic Lithiasis Complicated byImagologically Diagnosed MassBeihui Xue1,Sunjie Wu1,Minghua Zheng2,Huanchang Jiang3,Jun Chen3,Zhenghao Jiang3,Tian Tian3,Yifan Tu3,Huanhu Zhao4,Xia
2、n Shen5,Kuvaneshan Ramen6,Xiuling Wu7,Qiyu Zhang8,Qiqiang Zeng5*and Xiangwu Zheng1*1Radiological Department,The First Affiliated Hospital of Wenzhou Medical University,Wenzhou,China,2Department ofHepatology,The First Affiliated Hospital of Wenzhou Medical University,Wenzhou,China,3The First Clinical
3、 Medical Collegeof Wenzhou Medical University,Wenzhou,China,4School of Pharmacy,Minzu University of China,Beijing,China,5TheSecond Affiliated Hospital and Yuying Childrens Hospital of Wenzhou Medical University,Wenzhou,China,6Dr A.G JeetooHospital,Port Louis,Mauritius,7Department of Pathology,The Fi
4、rst Affiliated Hospital of Wenzhou Medical University,Wenzhou,China,8Department of Hepatobiliary Surgery,The First Affiliated Hospital of Wenzhou Medical University,Wenzhou,ChinaBackground:This study was conducted with the intent to develop and validate aradiomic model capable of predicting intrahep
5、atic cholangiocarcinoma(ICC)in patientswith intrahepatic lithiasis(IHL)complicated by imagologically diagnosed mass(IM).Methods:A radiomic model was developed in a training cohort of 96 patients with IHL-IMfrom January 2005 to July 2019.Radiomic characteristics were obtained from arterial-phase comp
6、uted tomography(CT)scans.The radiomic score(rad-score),based onradiomic features,was built by logistic regression after using the least absolute shrinkageand selection operator(LASSO)method.The rad-score and other independent predictorswere incorporated into a novel comprehensive model.The performan
7、ce of the Model wasdetermined by its discrimination,calibration,and clinical usefulness.This model wasexternally validated in 35 consecutive patients.Results:The rad-score was able to discriminate ICC from IHL in both the training group(AUC 0.829,sensitivity 0.868,specificity 0.635,and accuracy 0.72
8、3)and the validationgroup(AUC 0.879,sensitivity 0.824,specificity 0.778,and accuracy 0.800).Furthermore,the comprehensive model that combined rad-score and clinical features was great inpredicting IHL-ICC(AUC 0.902,sensitivity 0.771,specificity 0.923,and accuracy 0.862).Conclusions:The radiomic-base
9、d model holds promise as a novel and accurate tool forpredicting IHL-ICC,which can identify lesions in IHL timely for hepatectomy or avoidunnecessary surgical resection.Keywords:intrahepatic cholangiocarcinoma,intrahepatic lithiasis,radiomics,risk factors,nomogramFrontiers in Oncology|www.frontiersi
10、n.orgJanuary 2021|Volume 10|Article 5982531Edited by:Romain-David Seban,Institut Curie,FranceReviewed by:Pierre Decazes,Centre Henri Becquerel Rouen,FranceZhongxiang Ding,Zhejiang University,China*Correspondence:Qiqiang ZXiangwu ZhengXiangwuZSpecialty section:This article was submitted toCancer Imag
11、ing andImage-directed Interventions,a section of the journalFrontiers in OncologyReceived:24 August 2020Accepted:11 November 2020Published:07 January 2021Citation:Xue B,Wu S,Zheng M,Jiang H,Chen J,Jiang Z,Tian T,Tu Y,Zhao H,Shen X,Ramen K,Wu X,Zhang Q,Zeng Q andZheng X(2021)Development andValidation
12、 of a Radiomic-Based Modelfor Prediction of IntrahepaticCholangiocarcinoma in Patients WithIntrahepatic Lithiasis Complicated byImagologically Diagnosed Mass.Front.Oncol.10:598253.doi:10.3389/fonc.2020.598253ORIGINAL RESEARCHpublished:07 January 2021doi:10.3389/fonc.2020.598253INTRODUCTIONIntrahepat
13、ic cholangiocarcinoma(ICC)is the second mostprevalent liver malignancy following hepatocellular carcinoma,and its global disease incidence is increasing(1,2).The riskfactors for ICC are complex,but recently intrahepatic lithiasis(IHL)has been confirmed as a strong risk factor.High Oddsratios(ORs)hav
14、e been found for developing ICC due tohepatolithiasis,up to 50 in Korea(3),six in China(4),andseven in Italy(5).Studies have reported that about 2.3 to13.0%of patients with hepatolithiasis end up developingcholangiocarcinoma(611),and 6570%of patients in Taiwanwho underwent resection for cholangiocar
15、cinoma suffer fromconcomitant hepatolithiasis(12,13).It is very difficult for a clinical surgeon to identify ICC early inpatients with IHL because there are no specific symptoms andradiologicalfeatures.Althoughtissuebiopsycanbeusedtoconfirma histological diagnosis,it is not routinely recommended in
16、ICC(14),especially in IHL-ICC where negative biopsy results do notexclude ICC given the significant potential for sampling error.Thepreoperative diagnosis for IHL-ICC is mainly obtained from acombination of imaging,serum carcinoembryonic antigen(CEA),and cancer antigen 19-9(CA 19-9).However,the curr
17、entdiagnostic accuracy of IHL-ICC is low,generally ranging from 30to 65%(7,10,11,15,16).Recently,we have increased thediagnostic accuracy to 78.5%through developing a nomogramfor patients with IHL complicated by imagologically diagnosedmass(17).Despite this improvement,the accuracy of preoperativeim
18、aging diagnosis in the nomogram was still low because it wasperformed by two radiologists based on their experience.In recentyears,radiomics has been introduced in clinic to identify livertumors(18);however,no studies have used the radiomic approachfor diagnosing IHL-ICC.Therefore,there is an urgent
19、 need todevelop a radiomic model capable of improving the diagnosticaccuracy of IHL-ICC.In this study,we aimed to identify the radiomic features ofIHL-ICC,develop a predictive model that combined radiomicscore(rad-score)and clinical features for preoperative identificationof ICC among patients with
20、IHL,and also to validate its predictivecapacity in an independent data sets.PATIENTS AND METHODSPatients SelectionAll patients involved in this retrospective study that constitutedthe training cohort were diagnosed with intrahepatic lithiasis(IHL)complicated by imagologically diagnosed mass(IM)(IHL-
21、IM)andunderwenthepatectomyatTheFirstAffiliatedHospitalofWenzhou Medical University(WMU)from January 2005 to July2019.The database from our hospital was screened meticulouslyto select the potentially eligible patients who were;(1)withpathological diagnosis of ICC or IHL and(2)with availablehigh-quali
22、ty contrast-enhanced computed tomography(CT)before surgical resection.The clinical characteristics of thesequalified patients were recorded.This retrospective study wasreviewed and approved by the Institutional Review Board(IRB)ofthe First Affiliated Hospital of WMU,and a waiver of writteninformed c
23、onsent was granted by the IRB due to the retrospectivenature of this study in which de-identified data were usedand analyzed.The patients for the external validation cohort were selectedfrom the Second Affiliated Hospital of WMU,whose IRBapproved the validation study.Details for the recruitment and
24、selection criteria of thepatients included in this study were shown in Figure 1.CT Image Acquirement,TumorSegmentation,and Radiomic FeatureExtractionAll patients were assessed with contrast-enhanced CT using theLifeX software tools(19).Two radiologists(BX and SW)whowere blinded to the pathologic det
25、ails,reviewed transverse CTFIGURE 1|Proceeding flow of enrollment.Xue et al.Radiomic Model to Predict CholangiocarcinomaFrontiers in Oncology|www.frontiersin.orgJanuary 2021|Volume 10|Article 5982532images to determine respectively the features of the mass locationand boundary.The radiomic workflow
26、is depicted in Figure 2.Image featureextraction was performed on retrieved arterial phase CT images(5 mm thickness).The pre-processing procedure i.e.,theuniform of window width(200 Hu),window level(45 Hu),and pixel size(512 512)was undertaken before featureextraction.Manual segmentation of tumor reg
27、ions of interest(ROI)was carried out by two different radiologists(BX and SW).Each transverse slice consisted of cuts made along the primarytumor contour.A total of fifty-two quantified texture featureswere extracted,including features from histogram-based matrixand shape-based matrix from the first
28、 order and features fromgray-level co-occurrence matrix(GLCM),gray-level zone lengthmatrix(GLZLM),neighborhood gray-level dependence matrix(NGLDM),and gray-level run length matrix(GLRLM)fromsecond or higher order(20).A detailed description of all thesecharacteristics can be found in https:/www.lifex
29、soft.org/index.php/resources/19-texture/radiomic-features.All original dataabout extracted features are displayed in the SupplementaryMaterial 1 and Supplementary Material 2.Radiomic Feature Selection and SignatureConstructionWe devised a two-step procedure for dimensionality reductionand selection
30、of robust features.Firstly,we calculated theintraobserver and interobserver reliability for each ROI basedradiomic feature,extracted from 50 randomly chosen images.Toassess interobserver reliability,the ROI segmentation wasperformed by two experts one radiologist(reader 1,BX)andone hepatobiliary sur
31、geon(reader 2,QZ)who were blinded toboth the clinical and pathologic details.To evaluate intraobserverreliability,reader 1 repeated the ROI segmentation and featureextraction procedure twice over a period of one month.Thereliability was calculated byusing intraclass correlation coefficient.Radiomic
32、features with both intraobserver and interobserverintraclass correlation coefficient values greater than 0.55(demonstrating at least moderate stability)were selected forsubsequent investigation.Secondly,the least absolute shrinkageand selection operator(LASSO)logistic regression algorithm wasapplied
33、 to the training cohort in order to determine which ICC-related features had non-zero coefficients while being cross-validated 10 times by the penalty parameter.A radiomicsignature was generated via a linear combination of selectedfeatures weighted by their respective coefficients(21).Development,Pe
34、rformance,and Validationof a Radiomic NomogramA radiomic model incorporating the radiomic signature,as wellas independent risk factors that were obtained in our previousresearch for IHL-ICC(17),was constructed based on the resultsof the multivariate logistic regression analysis performed on thetrain
35、ing cohort.A radiomic nomogram was then constructed inorder to provide clinicians with a visual tool through the use ofthe selected covariates.Furthermore,a clinical model based onmultivariate logistic regression analysis of candidate predictors,with the exception of radiomic signature,was developed
36、.Wecalculated the area under the curve(AUC)of the receiveroperating characteristic curve(ROC)to measure the discriminationperformance of established models,and through the use of theDeLong algorithm(22),we compared the differences in AUCestimates between the various models.Calibration curves weregra
37、phed,through bootstrapping(resampled 1,000 times),toevaluate the predictive accuracy of the radiomic nomogram,ABDECFIGURE 2|Workflow of required steps in this current study.(A)Manual segmentation on arterial phase CT scans;(B)Quantification of tumor intensity,shape,andtexture through radiomic featur
38、es collected by LIFEx software from inside the defined tumor contours on CT images.(C)For feature selection,two successive stepsare the reliability assessment regarding the extracted features,followed by the LASSO method.A radiomic signature was obtained by combining the selectedfeatures by their re
39、spective coefficients,linearly.(D)By measuring the area under a receiver operating characteristic(ROC)curve and the calibration curve,theperformance of the prediction model can be analyzed.(E)A radiomic nomogram was built in order to provide clinicians with a visual tool through the use of theselect
40、ed covariates,followed by decision curve.Xue et al.Radiomic Model to Predict CholangiocarcinomaFrontiers in Oncology|www.frontiersin.orgJanuary 2021|Volume 10|Article 5982533followed by a HosmerLemeshow test(P 0.05 indicating good fit)(23).The performance of the radiomic model was then externallytes
41、ted through an independent validation cohort.Clinical Utility of the Radiomic NomogramThe net benefits at different threshold probabilities werequantified by a decision curve analysis(DCA)(24),therebyestimating the clinical utility of the established models in thevalidation cohort.Statistical Analys
42、isNumerical variables were compared by means of the t-test orMannWhitney U test,and categorical variables were comparedusing the c2 test or Fishers exact test,where appropriate.Univariate and multivariate Cox regression analyses wereperformed to determine predictors of IHL-ICC.All variableswith a p-
43、value 0.05 in univariate analysis were selected formultivariate analysis.Statistical analyses were performed withthe R software(version 3.4.4,http:/www.R-project.org),theEmpowerStats software(,X&Ysolutions,Inc.Boston MA).The R package“glmnet”was usedto perform LASSO binary logistic regression analys
44、is;the“rms”package,to create the nomogram;and the“pROC”package,toanalyze ROC curves.A two-sided p-value 5mg/L)+0.64579*(if143.15U/mlCA 19-937 U/ml)+1.56721*(if CA 19-9143.15 U/ml),and presented as a nomogram(Figure 3B).The model is capable ofindicating a good fit,as proved the HosmerLemeshow test(p=
45、0.764),and the calibration of the nomogram was likewise well-calibrated,as illustrated in Figure 2D.In the training cohort,thecomprehensive model displayed the highest discrimination betweenIHL-ICC and IHL-IBI with an AUC of 0.908(95%CI:0.833,0.970)(sensitivity0.771,specificity0.923,andaccuracy0.862
46、);thedetectedAUC value was higher than that of the radiomic signature model(AUC,0.829;p 0.05)and clinical prediction model and(AUC,0.838;p 0.05)(Figure 3C).In the validation cohort,thecomprehensive model presented the greatest AUC(0.879;95%CI:0.768,0.990)as well,which confirms that the comprehensive
47、model is capable of better predictive efficacy than either theradiomic signature model(AUC,0.824;p 0.05)or clinicalprediction model alone(AUC,0.755;p 0.05)(Figure 3D).Clinical UseThe DCA for the radiomic nomogram,the clinical predictionmodel,and the comprehensive model are presented in Figure 4.The
48、comprehensive model is capable of providing a better netbenefit when predicting ICC in IHL patients,when comparedwith the other two models(demonstrated by the thresholdprobabilities of more than 10%),and particularly,in situationswhere there is no alternative prediction model available.DISCUSSIONThe
49、 accurate diagnosis for IHL patients with ICC is extremelyimportant because it can facilitate the decision making withregard to surgical treatment at an early stage.The present work isthe first attempt to propose a comprehensive model combinedwith radiomic and clinical signatures that can improve th
50、eXue et al.Radiomic Model to Predict CholangiocarcinomaFrontiers in Oncology|www.frontiersin.orgJanuary 2021|Volume 10|Article 5982534current diagnostic accuracy standard of ICC in patients with IHL.The prediction model was validated internally and externally.In a recent study,we had developed a nom