Ultrasound-based Radiomics Signature: A Potential Biomarker For Predicting Biological Characteristics Of Hepatocellular Carcinoma

Objectives: the purpose of this study was to establish and validate a prediction model based on the radiomics features of ultrasound to evaluate the biological characteristics of Hepatocellular carcinoma(HCC)patients in a noninvasive manner.Methods: this study included 414 patients with single-focus HCC.We focused on the evaluation of eight biological characteristics: microvascular invasion,satellite nodules,glypican-3,Ki67,p53,vascular endothelial growth factor(VEGF),NM23,and cytokeratin 7(CK7).The region of interest(ROI)for HCC was identified by manually mapping the tumor profile on grayscale ultrasound images.We extracted the radiomics features from the ultrasonic images.Then,the dimensionality reduction method was used to select the radiomics features used to evaluate the 9 biological characteristics of HCC,and the classifier was used to establish the radiomics prediction models.Finally,independent validation data sets were used to evaluate the performance of the radiomics prediction model.The area under curve(AUC)of the receiver operating characteristic curve(ROC)was used to evaluate the predictive ability of the model.Results: We extracted 5234 radiomics features based on ultrasound images.Based on the extracted features of radiomics,we established the best radiomics prediction model for each subgroup.Through the feature selection method +classifier combination: Spearman + statistical test + Random Forest + Bagging,the best radiomics prediction model for evaluating microvascular invasion characteristics was obtained.Satellite nodule subgroup to obtain the best radiomics prediction model feature selection method + classifier combination is Spearman + statistical test + recursive feature elimination based on support vector machine(SVM-RFE)+ decision tree.The Glypican-3 subgroup was Spearman +statistical test + Random Forest + Random Forest.The Ki67 subgroup was Spearman + SVM-RFE + Bagging.The p53 subgroup was Spearman + Statistical Test+ Random Forest +Random Forest.The VEGF subgroup was Spearman + Statistical Test + SVM-RFE + Multi-layer Perception.The NM23 subgroup was Spearman + Statistical Test + Random Forest + support vector machine(SVM).CK7 subgroup consisted of Spearman+ least absolute shrinkage and selection operator(LASSO)+ random forest.The AUC test and validation sets of these predictive models were 0.87 and0.80(microvascular invasion),0.98 and 0.77(satellite nodule),0.95 and 0.67(Glypican-3),0.97 and 0.69(Ki67),0.93 and 0.75(p53),0.97 and 0.62(VEGF),0.94 and 0.62(NM23)and 0.77 and 0.73(CK7),respectively.The models established in the 8 subgroups were able to distinguish the positive and negative expressions of the biological characteristics of HCC in the Training set,and had a good ability to identify the microvascular invasion,satellite nodule,p53 and CK7 subgroups in the validation set.Conclusion: The radiomics models based on ultrasound images in this study can effectively predict preoperative microvascular invasion,satellite nodule,P53 and CK7 in HCC patients.It provides a new perspective for preoperative assessment of the biological characteristics of HCC patients,and then provide guidance for the diagnosis,treatment and prognosis of HCC patients.