CT-based Radiomics For Pretreatment Prediction Of Microvascular Invasion Risk And Prognosis In Primary Hepatocelluar Carcinoma

Back groundSolid tumors with spatially and temporally heterogeneity,limiting the molecular detection based on invasive biopsy,but this aspect of medical imaging technology has huge potential advantage,which can capture the intra-tumor heterogeneity in a non-invasive method.In the past few decades,medical imaging technology has been innovating,new imaging agents and standardized protocols have enabled quantitative imaging to solve the problem.Therefore,it is necessary to automatically and repeatably extract more image information from the images.Radiomics——the high throughput extraction of group characteristics from medical image images is a method to solve this problem and is full of promise,but it needs to be further verified by multiple centers.It is worth noting that the radiomics signature has been verified in multiple tumors and is closely related to gene mutation,prognosis and curative effect.It suggests a new question:can we accurately predict the occurrence of micro vascular invasion of liver cancer before surgery?Is it possible to predict the prognosis before partial resection of liver cancer?PurposeFirst,to develop and validate a radiomics nomogram for preoperative prediction of microvascular invasion(MVI)in hepatitis B virus(HBV)-related hepatocellular carcinoma(HCC),so as to provide guidance for clinical treatment decisions.Second,to explore the feasibility of establishing a prognostic prediction model based on preoperative CT radiomics,so as to provide new ideas for precise treatment of liver cancer.MethodsA total of 304 eligible patients with HCC were randomly divided into the training(n =184)and independent validation(n = 120)cohorts.Portal venous and arterial phase computed tomography data of the HCCs were collected for extraction of radiomic features by MATAB2014b.Using the least absolute shrinkage and selection operator algorithm,the training set was processed for reducingdata dimension,feature selection,and construction of a radiomics signature.Then,aprediction model including the radiomics signature,radiologic features,and alpha-fetoprotein(AFP)level as presented in a radiomics nomogram was developed using multivariable logistic regression analysis.The radiomics nomogram was analyzed in terms of its discrimination ability,calibration and clinical usefulness.The internal cohort data were validated using the radiomics nomogram.On the other hand,one hundred and seventy-seven patients with hepatocellular carcinoma were randomly divided into a training cohort(n = 113)and an independent validation cohort(n = 64).Portal-venous and arterial phase computed tomography(CT)of HCC was gathered to extract radiomic features by MATAB2014b.Least absolute shrinkage and selection operator(LASSO)algorithm was used to select optimal features and build a radiomics signature in the training set.Multivariable regression analysis was then used to construct a radiomics nomogram.The performance of the radiomics nomogram was then estimated by discrimination,calibration and clinical usefulness.The internal cohort was validated.ResultsA total of 8 features were chosen from 980 radiomics features,and radiomics signature was established.The radiomics signature was significantly associated with MVI status(P<0.001,both cohorts).Predictors including the radiomics signature,nonsmooth tumor margin,hypoattenuating halos,internal arteries,and alpha-fetoprotein level were reserved in the individualized prediction nomogram.The model exhibited good calibration and discrimination in the training and validation cohorts(C-index[95%confidence interval]:0.846[0.787-0.905]and 0.844[0.774-0.915],respectively).Its clinical usefulness was confirmed by the findings of decision curve analysis.In other study of liver resection,the radiomics signature included 19 selected features and was significantly associated with disease-free survival(DFS)(P<0.001 and P = 0.00013,respectively)and overall survival(OS)(both for P<0.0001)in both training and validation cohorts,who underwent partial hepatectomy.Combining the radiomics signature,Barcelona Clinic Liver Cancer(BCLC)stage and alpha-fetoprotein level,we developed the individualized radiomics nomogram.The model showed good discrimination,with a C-index of 0.782(95%CI:0.730-0.834),and good calibration.The radiomics nomogram also showed good discrimination of C-index 0.743(95%CI:0.657-0.829),and good calibration in the validation cohort.The clinical usefulness of the radiomics model was confirmed by decision curve analysis.ConclusionsWe first used the radiomics to predict the occurrence of MVI and the prognosis after hepatectomy.As a new preoperative prediction method,the nomogram model of the imaging group showed good prediction accuracy of MVI status in HBV-related hepatocellular carcinoma patients.On the other hand,we build the radiomics nomogram model before liver resection in patients with liver cancer to predict the DFS on the individualization,non-invasive assessment,which can help us to identify more benefit in surgical treatment of HCC patients,avoid excessive or insufficient treatment.Those radiomics studies provide a new method for the precise treatment of liver cancer in the future.