摘要
目的:探讨基于T2WI、DWI的影像组学鉴别肝细胞肝癌(HCC)与肝富血供良性病变(BLLs)的价值。方法:回顾性收集114例经病理证实为HCC(n=77)、血管瘤(n=23)、肝局灶性结节样增生(n=8)、肝血管平滑肌脂肪瘤(n=4)、肝脓肿(n=2)患者的临床与影像资料,所有患者均在术前行3.0T MRI增强检查。利用ITK-SNAP软件勾画病灶,通过最小绝对收缩和选择算子回归以及最小冗余最大相关筛选影像组学特征,计算影像组学分数(Rad-score),使用多元logistic回归算法构建预测模型。绘制受试者工作特性(ROC)曲线评估预测模型的效能,并与两位不同年资影像医生的诊断结果进行比较。结果:筛选得到5个T2WI特征、6个DWI特征,计算Rad-score,结合3个临床独立预测因子(年龄、性别、病灶位置)用于建立诺模图。与临床预测模型、影像组学模型相比,诺模图模型在训练集、测试集中显示出较高的诊断效能,AUC分别为0.988、0.955,敏感度分别为0.981、0.958,特异度分别为0.923、0.727,准确度分别为0.962、0.886,其诊断效能显著优于医生1(AUC=0.808,P<0.001)及医生2(AUC=0.780,P<0.001)。结论:基于MRI多模态影像组学的诺模图模型在鉴别HCC与BLLs方面明显优于单一预测模型和传统影像诊断,可作为临床诊断的辅助工具。
Objective:To explore the value of radiomics based on T 2WI,DWI in differentiating hepatocellular carcinoma(HCC)from blood-rich benign liver lesions(BLLs).Methods:The clinical and imaging data of 114 patients with HCC(n=77),hemangioma(n=23),hepatic focal nodular hyperplasia(n=8),hepatic angiomyolipoma(n=4)and hepatic abscess(n=2)confirmed by pathology were retrospectively collected.All patients underwent 3.0T MRI enhanced examination before surgery.The lesions were delineated by ITK-SNAP software,and the radiomics features were screened by minimum absolute contraction and selection operator regression and minimum redundant maximum correlation and the radiomics score was calculated.Multivariate Logistic regression algorithm was used to construct prediction models.The efficacy of prediction models was evaluated by receiver operating characteristic(ROC)curves and compared with the results of two radiologists with different seniority.Results:Five T 2WI features and six DWI features were screened,and rad-score was calculated,which was combined with three independent clinical predictors(age,sex and location of lesions)to construct a nomogram model.Compared with clinical prediction model and radiomics model,the nomogram model showed higher diagnostic efficiency in training set and test set,with AUC of 0.988 and 0.955,sensitivity of 0.981 and 0.958,specificity of 0.923 and 0.727,accuracy of 0.962 and 0.886,respectively.The efficacy was significantly better than that of doctor 1(AUC=0.808,P<0.001)and doctor 2(AUC=0.780,P<0.001).Conclusions:The nomogram model based on multimodal MRI radiomics is superior to the single prediction model and traditional imaging diagnosis in distinguishing HCC from blood-rich benign liver lesions,and can be used as an auxiliary tool for clinical diagnosis.
作者
杨晨
朱帆东
夏阳
楼天奇
张敏鸣
赵振华
MRI YANG Chen;ZHU Fan-dong;XIA Yang(Department of Radiology,Shaoxing Hospital,Zhejiang University,Zhejiang 312000,China)
出处
《放射学实践》
CSCD
北大核心
2023年第5期581-586,共6页
Radiologic Practice
基金
浙江省卫生健康科技计划(2022KY1321)
浙江省医药卫生厅面上项目(2020KY323)
绍兴市人民医院青年科研基金项目(2021YB05、2021YB06)
绍兴市肿瘤功能分子成像与介入诊疗重点实验室资助。
关键词
肝细胞肝癌
影像组学
富血供病变
磁共振成像
扩散加权成像
Hepatocellular carcinoma
Radiomics
Blood-rich lesions
Magnetic resonance imaging
Diffusion weighted imaging
