机器学习影像组学可预测周围肝胆管癌根治性
2021-8-17 来源:本站原创 浏览次数:次白癜风权威医院 http://yyk.39.net/bj/zhuanke/89ac7.html
Radiomics
Abstract
文末有中文翻译(有道)Backgroundandaims:Upto40%-65%ofpatientswithperihilarcholangiocarcinoma(PHC)rapidlyprogresstoearlyrecurrence(ER)evenaftercurativeresection.QuantificationofERriskisdifficultandareliableprognosticpredictiontoolisabsent.Wedevelopedandvalidatedamultilevelmodel,integratingclinicopathology,molecularpathologyandradiology,especiallyradiomicscoupledwithmachine-learningalgorithms,topredicttheERofpatientsaftercurativeresectioninPHC.
Methods:Intotal,patientswhounderwentcontrast-enhancedCT(CECT)andcurativeresectionat2-institutionswereretrospectivelyidentifiedandrandomlydividedintotraining(n=),internalvalidation(n=70),andexternalvalidation(n=37)sets.Amachine-learninganalysisof18,radiomicfeaturesbasedonmulti-phaseCECTand48clinico-radiologiccharacteristicswasperformedforthemultilevelmodel.
Results:Comprehensively,7independentfactors(tumordifferentiation,lymphnodemetastasis,preoperativeCA19-9level,enhancementpattern,A-Shrinkscore,V-Shrinkscore,P-Shrinkscore)werebuilttothemultilevelmodelandquantifiedtheriskofER.Webenchmarkedthegainindiscriminationwiththeareaunderthecurve(AUC)of0.,superiortotherivalclinicalandradiomicmodels(AUCs0.-0.).Theaccuracy(ACC)ofthemultilevelmodelwas0.,whichwassignificantlyhigherthanthoseoftheconventionalstagingsystems(AJCC8th(0.),MSKCC(0.),andGazzaniga(0.)).
Conclusion:Theradiomics-basedmultilevelmodeldemonstratedsuperiorperformancetorivalmodelsandconventionalstagingsystems,andcouldserveasavisualprognostictooltoplansurveillanceofERandguidepostoperativeindividualizedmanagementinPHC.
Keywords:earlyrecurrence;machinelearning;multilevelmodel;perihilarcholangiocarcinoma;radiomics.
影像组学中文摘要
背景与目的:即使根治性切除术后,多达40%-65%的肝门周围胆管癌(PHC)患者也迅速发展为早期复发(ER)。ER风险的量化是困难的,并且缺乏可靠的预后预测工具。我们开发并验证了一个多级模型,该模型整合了临床病理学,分子病理学和放射学,尤其是放射线学与机器学习算法,可以预测PHC根治性切除术后患者的ER。
方法:回顾性分析例接受2机构造影对比CT(CECT)和根治性切除术的患者,并将其随机分为训练(n=),内部验证(n=70)和外部验证(n=37)套。对多级模型进行了基于多阶段CECT的18个放射学特征和48个临床放射学特征的机器学习分析。
结果:全面建立了7个独立因素(肿瘤分化,淋巴结转移,术前CA19-9水平,增强模式,A-Shrink评分,V-Shrink评分,P-Shrink评分),并量化了ER患病风险。我们以曲线下面积(AUC)为0.作为鉴别增益的基准,优于竞争对手的临床和放射模型(AUC0.-0.)。多级模型的准确性(ACC)为0.,明显高于常规分期系统(AJCC第八(0.),MSKCC(0.)和Gazzaniga(0.))。
结论:基于放射学的多级模型表现出优于竞争对手模型和常规分期系统的性能,并可作为视觉预后工具来计划ER的监测并指导PHC的术后个体化管理。
关键字:早期复发;机器学习;多层次模型;肝门周围胆管癌;影像组学
原文链接: