Radiomic analysis of Gd-EOB-DTPA-enhanced MRI predicts Ki-67 expression in hepatocellular carcinoma

Abstract

Background Nuclear protein Ki-67 indicates the status of cell proliferation and has been regarded as an attractive biomarker for the prognosis of HCC. The aim of this study is to investigate which radiomics model derived from different sequences and phases of gadolinium-ethoxybenzyl-diethylenetriamine pentaacetic acid (Gd-EOB-DTPA)-enhanced MRI was superior to predict Ki-67 expression in hepatocellular carcinoma (HCC), then further to validate the optimal model for preoperative prediction of Ki-67 expression in HCC. Methods This retrospective study included 151 (training cohort: n = 103; validation cohort: n = 48) pathologically confirmed HCC patients. Radiomics features were extracted from the artery phase (AP), portal venous phase (PVP), hepatobiliary phase (HBP), and T2-weighted (T2W) images. A logistic regression with the least absolute shrinkage and selection operator (LASSO) regularization was used to select features to build a radiomics score (Rad-score). A final combined model including the optimal Rad-score and clinical risk factors was established. Receiver operating characteristic (ROC) curve analysis, Delong test and calibration curve were used to assess the predictive performance of the combined model. Decision cure analysis (DCA) was used to evaluate the clinical utility. Results The AP radiomics model with higher decision curve indicating added more net benefit, gave a better predictive performance than the HBP and T2W radiomic models. The combined model (AUC = 0.922 vs. 0.863) including AP Rad-score and serum AFP levels improved the predictive performance more than the AP radiomics model (AUC = 0.873 vs. 0.813) in the training and validation cohort. Calibration curve of the combined model showed a good agreement between the predicted and the actual probability. DCA of the validation cohort revealed that at a range threshold probability of 30–60%, the combined model added more net benefit compared with the AP radiomics model. Conclusions A combined model including AP Rad-score and serum AFP levels based on enhanced MRI can preoperatively predict Ki-67 expression in HCC.

A Potential Prognostic Marker for Recognizing VEGF-Positive Hepatocellular Carcinoma Based on Magnetic Resonance Radiomics Signature

“…ICC test: The reproducible features were set as the features with an intra- and inter-observer ICC greater than 0.8 ( 27 ). These robust features were applied in the following feature selection.…”

Progress of MRI Radiomics in Hepatocellular Carcinoma

“…To date, radiogenomics studies have described the semantic features obtained from MRI (112)(113)(114)(115). MRI radiogenomics has the value of predicting gene features with prognostic and therapeutic significance (16,17,(116)(117)(118)(119)(120)(121)(122)(123). Taouli et al (113) found that there was a strong connection between imaging features, such as the "infiltrative pattern", "mosaic appearance", and "presence of macrovascular invasion", and an aggressive genomic signature determined previously.…”

Added Value of Viscoelasticity for MRI-Based Prediction of Ki-67 Expression of Hepatocellular Carcinoma Using a Deep Learning Combined Radiomics (DLCR) Model

“…Hu et al [ 6 ] reported that histogram-based parameters from apparent diffusion coefficient (ADC) maps and arterial phase (AP) images could be used to determine the Ki-67 labeling index (LI) in HCC with an AUC up to 0.826. Similarly, Fan et al [ 7 ] proposed a combined model including an arterial phase-based RAD-score and serum alpha-fetoprotein (AFP) level to predict Ki-67 expression preoperatively. The combined model (AUC, 0.863), that included the AP RAD-score and serum AFP level, demonstrated improved performance when compared with the single AP radiomics model (AUC, 0.813).…”