基于超声影像组学的原发性肝细胞癌分级预测

Prediction of grade of hepatocellular carcinoma by radiomics based on ultrasound

目的:针对原发性肝细胞癌(HCC)肿瘤分级预测难题,提出一种基于灰阶超声成像的影像组学预测模型。方法:首先,由超声医生对肿瘤区域进行手动分割,其次,采用影像组学方法对肿瘤区域提取形状、一阶统计、纹理特征,计算特征间Pearson相关系数剔除冗余特征,最后通过单变量分析筛选得到特征子集,采用LASSO构建HCC分级预测模型;利用留一法计算模型的受试者操作特性曲线下的面积(AUC)评估模型对HCC分级的预测能力。结果:利用43例经手术病理证实的HCC患者的灰阶超声图像构建HCC分级预测模型,所建模型由6个与分级高度相关的影像特征组成,模型具有较强的预测能力(AUC=0.76)。结论:基于灰阶超声成像的影像特征与HCC分级高度相关,所建影像组学模型能够较好地预测HCC分级。

Objective To propose a radiomics model based on gray-scale ultrasound images for solving the problem of predicting the grade of hepatocellular carcinoma(HCC).Methods Firstly,the tumor areas were segmented by an ultrasound physician,and then various features of tumor areas,including shape,the first order statistical properties and texture features were extracted by radiomics.Pearson's correlation coefficient was used to eliminate the redundant features.Finally,univariate analysis was used for obtaining the optimal feature subset,and LASSO for constructing a model for predicting the grade of HCC.The area under the receiver operating characteristic curve(AUC)of the model was calculated by leave-one-of-cross validation so as to evaluate the prediction ability of the model.Results The radiomics model for prediction of the grade of HCC was constructed using grayscale ultrasound images of 43 cases of HCC confirmed by operation and pathology.The obtained model was composed of 6 image features which was highly correlated with grading,and the results showed that the proposed model had preferable predication performances(AUC=0.76).Conclusion The image features based on gray-scale ultrasound images are highly correlated with the grade of HCC.The established radiomics model can be used to better predict the grade of HCC.