摘要
目的探究能谱CT成像及其影像组学对肺部实性病变性质的鉴别诊断效能,并与临床危险因素、CT主观征象、能谱CT定量参数对照。方法对194例肺部实性病变的能谱CT图像进行回顾性研究,分成训练组(n=155)和验证组(n=39),基于动脉期的碘基物质图提取影像组学特征;采用"逻辑回归"分类器模型,结合碘浓度(IC)及标准化碘浓度(NIC)构建能谱CT定量参数模型,结合患者临床危险因素、CT主观征象构建临床模型,结合IC、NIC及所提取的影像组学特征构建联合模型;受试者工作特征曲线(ROC)及曲线下面积(AUC)用于评价模型的鉴别诊断效能,Delong检验用于比较模型间AUC差异是否具有统计学意义。结果能谱CT定量参数模型由IC、NIC构成,临床模型由年龄、性别、基础病史、病灶形态、分叶征、毛刺征、胸膜凹陷征、血管集束征、钙化构成;联合模型由IC、NIC与筛选出的9个影像组学特征构成;能谱CT定量参数模型、临床模型以及联合模型在训练组中的AUC分别为0.659(95%CI:0.575~0.736)、0.742(95%CI:0.663~0.812)、0.937(95%CI:0.884~0.971),在验证组中的AUC分别为0.665(95%CI:0.489~0.813)、0.777(95%CI:0.608~0.898)、0.912(95%CI:0.769~0.980),经过Delong检验,三种模型在训练组和验证组中的AUC差异均具有统计学意义(P<0.05)。结论能谱CT成像及其影像组学对肺部良恶性病变具有很高的鉴别诊断能力,明显高于临床危险因素、CT主观征象及能谱CT定量参数。
Objective The purpose of this study was to explore the ability of spectral CT-based imaging and radiomics features in the differential diagnosis of lung solid lesions, which was compared with clinical risk factors, CT signs and quantitative parameters of spectral CT. Methods 194 cases with pulmonary solid lesions, confirmed by pathology, were studied retrospectively.They were divided into training cohort(n=155)and validation cohort(n=39).Based on the iodine-based material decomposition images of spectral CT in arterial phase, the radiomics features were selected.The logistic regression classifier model is adopted.A quantitative parameter model of spectral CT was constructed via combining with iodine concentration(IC) and standardized iodine concentration(NIC).The clinical model was constructed by combining the clinical risk factors and CT subjective signs of patients.The combined model was constructed by combining IC,NIC and the radiomics characteristics extracted.Receiver operating characteristic curve(ROC) and area under the curve(AUC)were used to evaluate the differential diagnosis efficiency of the models in the training cohort and validation cohort.Delong test was performed to compare whether the difference of the AUC among models have statistical significance in the two cohort. Results The quantitative parameter model of spectral CT contains IC and NIC.The clinical model contained patient’s age, gender, basic medical history, lesion morphology, lobulation sign, prickle sign, pleural indentation sign, vascular cluster sign and calcification.The combined model was consisted of IC,NIC and 9 selected radiomics features.In training cohort, The AUC of quantitative parameter model of spectral CT,clinical model and combined model were 0.659(95%CI:0.575-0.736),0.742(95%CI:0.663-0.812)and 0.937(95%CI:0.884-0.971),respectively.In the validation cohort, they were 0.665(95%CI:0.489-0.813),0.777(95%CI:0.608-0.898)and 0.912(95%CI:0.769-0.980),respectively.Via Delong test, there were significant differences in AUC among
作者
徐鹤
王小雷
杨昭
李伟
王效静
刘浩
王健
谢宗玉
XU He;WANG Xiaolei;YANG Zhao(Department of Radiology,The First Affiliated Hospital of Bengbu Medical College,Bengbu,Anhui Province 233004,P.R.China)
出处
《临床放射学杂志》
北大核心
2021年第8期1510-1515,共6页
Journal of Clinical Radiology
基金
安徽省高校自然科学研究重点项目(编号:KJ2019A0402)
2021年第二批安徽省中央引导地方科技发展资金项目(编号:2020b07030008)
浙江省卫生健康科技计划项目(编号:2021KY602)。
关键词
能谱CT成像
影像组学
肺部良恶性病变
Spectral CT imaging
Radiomics
Benign and malignant pulmonary lesions
