Clinical‐Radiologic Morphology‐Radiomics Model on Gadobenate Dimeglumine‐Enhanced MRI for Identification of Highly Aggressive Hepatocellular Carcinoma: Temporal Validation and Multiscanner Validation

Abstract

BackgroundHighly aggressive hepatocellular carcinoma (HCC) is characterized by high tumor recurrence and poor outcomes, but its definition and imaging characteristics have not been clearly described.PurposeTo develop and validate a fusion model on gadobenate dimeglumine‐enhanced MRI for identifying highly aggressive HCC.Study TypeRetrospective.Population341 patients (M/F = 294/47) with surgically resected HCC, divided into a training cohort (n = 177), temporal validation cohort (n = 77), and multiscanner validation cohort (n = 87).Field Strength/Sequence3T, dynamic contrast‐enhanced MRI with T1‐weighted volumetric interpolated breath‐hold examination gradient‐echo sequences, especially arterial phase (AP) and hepatobiliary phase (HBP, 80–100 min).AssessmentClinical factors and diagnosis assessment based on radiologic morphology characteristics associated with highly aggressive HCCs were evaluated. The radiomics signatures were extracted from AP and HBP. Multivariable logistic regression was performed to construct clinical‐radiologic morphology (CR) model and clinical‐radiologic morphology‐radiomics (CRR) model. A nomogram based on the optimal model was established. Early recurrence‐free survival (RFS) was evaluated in actual groups and risk groups calculated by the nomogram.Statistical TestsThe performance was evaluated by receiver operating characteristic curve (ROC) analysis, calibration curves analysis, and decision curves. Early RFS was evaluated by using Kaplan–Meier analysis. A P value <0.05 was considered statistically significant.ResultsThe CRR model incorporating corona enhancement, cloud‐like hyperintensity on HBP, and radiomics signatures showed the highest diagnostic performance. The area under the curves (AUCs) of CRR were significantly higher than those of the CR model (AUC = 0.883 vs. 0.815, respectively, for the training cohort), 0.874 vs. 0.769 for temporal validation, and 0.892 vs. 0.792 for multiscanner validation. In both actual and risk groups, highly and low aggressive HCCs showed statistically significant differences in early recurrence.Data ConclusionThe clinical‐radiologic morphology‐radiomics model on gadobenate dimeglumine‐enhanced MRI has potential to identify highly aggressive HCCs and non‐invasively obtain prognostic information.Level of Evidence4Technical EfficacyStage 2