CT-Based Radiomics for Non-Invasive Prediction of Ki-67 Expression in Hepatocellular Carcinoma
Meilong Wu1, Zhiyong Du1, Ying Xiao2
1Division of Hepatobiliary and Pancreas Surgery, Department of General Surgery, Shenzhen People's Hospital (The Second Clinical Medical College, Jinan University, The First Affiliated Hospital, Southern University of Science and Technology), Shenzhen, Guangdong, People's Republic of China.
Purpose:
By far, non-invasive methods assessing Ki-67 are still scarce. This study was performed to evaluate the capability of radiomics based on contrast-enhanced CT in predicting expression of Ki-67 in hepatocellular carcinoma (HCC).
Patients And Methods:
HCC patients who underwent curative hepatectomy were included. The optimal Ki-67 cutoff value for prognostic stratification was determined using maximum selection rank statistics. The Least Absolute Selection and Shrinkage Operator (LASSO) regression analysis was used for dimension reduction and data screening to obtain radiomics features. The radiomics model, clinical model and combined model integrating radiomics features and clinical indicators for predicting Ki-67 were constructed. The predictive efficacy among the models was compared using C-index, and further verified through DeLong test, decision curves and clinical impact curves.
Results:
The optimal cutoff value for Ki-67 was 0.25. A total of 2553 radiomics features were obtained and after stability testing (by intraclass correlation coefficient ≥0.75) and feature selection (LASSO), four radiomics features that were highly correlated and stable with the expression of Ki-67 were included in the model construction. Multivariate analyses revealed that alpha-fetoprotein and intratumoral necrosis or radiomics features were independent predictors of Ki-67. The clinical model, radiomics model and combined model were constructed, respectively. The C-indices of Ki-67 for clinical model, radiomics model and the combined model were 0.75, 0.82 and 0.88. DeLong test, decision curves and clinical impact curves further confirmed that the inclusion of radiomics features improved the predictive efficacy of the model.
Conclusion:
The comprehensive model based on contrast-enhanced CT radiomics could non-invasively and effectively predict the expression of Ki-67, suggesting its value in clinical decision-making.
