摘要
目的探讨基于扩散加权成像(DWI)的影像组学模型对肝细胞癌和肝血管瘤进行鉴别诊断的可行性。方法回顾性分析2017年1月至2019年1月在广州中医药大学第一附属医院经病理或临床诊断为肝细胞癌和肝血管瘤,且行MRI扫描获得表观扩散系数(ADC)图像的126例患者(肝细胞癌68例,肝血管瘤58例)资料。对ADC图像的病灶区进行手动分割得到134个病灶(肝细胞癌68个,肝血管瘤66个),提取960个特征参数。其中94个病灶作为训练组,40个作为验证组。利用Lasso回归对特征参数降维并构建预测模型,采用受试者工作特征曲线(ROC)评价诊断效能。在验证组中,两名影像科医师仅根据DWI序列图像鉴别诊断肝细胞癌和肝血管瘤,比较其与影像组学模型的诊断效能差异。结果降维后筛选出14个系数非零的特征参数。训练组模型准确率95.7%(90/94),敏感度93.2%(41/44),特异度98.0%(49/50),曲线下面积(AUC)为0.994(95%CI为0.985~1.000);验证组模型准确率90.0%(36/40),敏感度87.5%(21/24),特异度93.8%(15/16),AUC为0.964(95%CI为0.910~1.000)。在验证组中,两名影像科医师诊断的AUC分别为0.832(95%CI为0.717~0.948)和0.819(95%CI为0.680~0.959)。结论基于DWI的影像组学模型对鉴别肝细胞癌和肝血管瘤具有很大价值。
Objective To determine the feasibility of a diffusion-weighted imaging(DWI)radiomics model for the differential diagnosis of hepatocellular carcinoma(HCC)and hepatic haemangioma.Methods A retrospective analysis was performed of 126 patients(68 liver cancers and 58 haemangiomas)who were diagnosed with HCC or hepatic haemangioma that was pathologically or clinically confirmed at the First Affiliated Hospital of Guangzhou University of Chinese Medicine from January 2017 to January 2019,who had undergone preoperative abdominal MRI scanning to obtain apparent diffusion coefficient(ADC)images.A total of 134 lesions(68 liver cancers and 66 haemangiomas)were examined by manual segmentation of the lesion area on the ADC image,and 960 radiomics features were extracted in each lesion.The primary dataset consisted of 94 lesions,and independent validation was conducted for 40 lesions.By using Lasso regression,the dimensions of the radiomics features were reduced,and a prediction model was constructed.The diagnostic performance of the model was analysed by using receiver operating characteristic(ROC)curves.In the independent validation dataset,the diagnostic performance was compared against the radiomics model and the diagnostic readings of two radiologists.Results After dimension reduction,14 radiomics signatures were finally selected.The radiomics model showed accuracy of 95.7%(90/94),sensitivity of 93.2%(41/44),specificity of 98.0%(49/50),and area under the curve(AUC)of 0.994[95%confidence interval(CI):0.985 to 1.000]in primary dataset and accuracy of 90.0%(36/40),sensitivity of 87.5%(21/24),specificity of 93.8%(15/16),and AUC of 0.964(95%CI:0.910 to 1.000)in the independent validation dataset.In the same independent validation dataset,the two radiologists achieved AUCs of 0.832 and 0.819,respectively.Conclusion This radiomics model based on ADC mapping can help to identify HCC and hepatic haemangioma.
作者
王禹博
王顺涛
谢磊
余芬芬
李伊凡
张昌政
WANG Yubo;WANG Shuntao;XIE Lei(Department of Radiology,the First Affiliated Hospital of Guangzhou University of Chinese Medicine,Guangzhou 510405,P.R.China)
出处
《临床放射学杂志》
CSCD
北大核心
2020年第3期481-486,共6页
Journal of Clinical Radiology
关键词
肝细胞癌
肝血管瘤
扩散加权成像
影像组学
Hepatocellular carcinoma
Hepatic haemangioma
Diffusion weighted imaging
Radiomics
作者简介
通讯作者:张昌政
