基于超声成像提取影像组学特征构建Rad-score评估肝脏占位性病变良恶性的研究

Evaluation of the benign and malignant of liver space-occupying lesions based on the radiomics features extracted by ultrasound imaging to construct Rad-score

目的探讨基于超声影像组学方法构建影像组学标签(Rad-score)评估肝脏占位性病变良恶性的可行性。方法纳入该院2016年5月至2020年7月肝脏占位性病变患者79例,所有患者均经穿刺活检或手术切除组织行病理检测确定患者病灶良恶性。患者均于穿刺活检或者手术前1周行超声成像,由该科室高年资医师基于超声影像对病灶进行感兴趣区域(ROI)勾画,提取影像组学特征,将入组患者按照随机原则以7∶3比例分为训练集和验证集,以训练集患者是否为恶性肿瘤为研究目的进行影像组学特征去冗除杂,并构建Rad-score,最后采用受试者工作特征(ROC)曲线进行诊断效能评估。结果共计提取了943个影像组学特征,基于训练组最小冗余最大相关(mRMR)特征去冗除杂,保留了30个特征,继而采用套索算法(LASSO)进行Rad-score构建,最终筛选出6个特征。训练集和测试集中良恶性患者Rad-score差异有统计学意义(P=0.000 vs.0.016)。基于Rad-score评估训练集和测试集中肝脏良恶性病变的曲线下面积分别为0.854、0.841。结论基于超声成像的影像组学可协助临床评估肝脏占位性病变的良恶性。

Objective To explore the feasibility of constructing Rad-score based on the ultrasound radiomics to evaluate the benign and malignant liver space-occupying lesions.Methods A total of 79 patients with liver space-occupying lesions in the hospital from May 2016 to July 2020 were included.All patients were identified as benign or malignant by pathological examination of biopsy sampling or surgical resection of tissue.All patients underwent ultrasound imaging one week before puncture sampling or surgery.A senior physician in the department outlined the region of interest of the lesion based on the ultrasound images,and the radiomics software Artificial Intelligence Kit was used to extract the imaging radiomics.The enrolled patients were randomly divided into the training set and the validation set at a ratio of 7∶3.For the purpose of the study,whether the patients in the training set had liver malignant tumor or not,the radiomics features were removed for the redundancy and impurity,and the Rad-score was constructed.Finally,the receiver operating characteristic(ROC)curve was used to evaluate the diagnostic performance.Results A total of 943 radiomics features were extracted,and 30 features were retained based on the minimum redundancy maximum relevance(mRMR)features of the training set.Then,the LASSO algorithm was used to construct the radiomics label,and finally six features were selected.There were statistically significant differences in the Rad-score between the benign and malignant patients in the training set and the test set(P=0.001 vs.P=0.016).The AUC of the benign and malignant liver lesions in the training set and the test set based on Rad-score evaluation were 0.854 and 0.841.Conclusion The ultrasound radiomics based on the ultrasound imaging can assist in the clinical evaluation of benign and malignant liver space-occupying lesions.