Preoperative Evaluation of Gd-EOB-DTPA-Enhanced MRI Radiomics-Based Nomogram in Small Solitary Hepatocellular Carcinoma (<= 3 cm) With Microvascular Invasion: A Two-Center Study

Background Preoperative evaluation of microvascular invasion (MVI) in small solitary hepatocellular carcinoma (HCC; maximum lesion diameter <= 3 cm) is important for treatment decisions. Purpose To apply gadolinium ethoxybenzyl diethylenetriamine pentaacetic acid (Gd-EOB-DTPA)-enhanced MRI to develop and validate a nomogram for preoperative evaluation of MVI in small solitary HCC and to compare the effectiveness of radiomics evaluation models based on different volumes of interest (VOIs). Study Type Retrospective. Population A total of 196 patients include 62 MVI-positive and 134 MVI-negative patients were enrolled (training cohort, n = 105; testing cohort, n = 45; external validation cohort, n = 46). Field Strength/Sequence 3.0 T, fat suppressed fast-spin-echo T2-weighted and Gd-EOB-DTPA-enhanced T1-weighted magnetization-prepared rapid gradient-echo sequences. Assessment Radiomics features were extracted on T2-weighted, arterial phase (AP), and hepatobiliary phase (HBP) images from different VOIs (VOIintratumor and VOIintratumor+peritumor) and filtered by the least absolute shrinkage selection operator (LASSO) regression. From VOIintratumor and VOIintratumor+peritumor, eight radiomics models were constructed based on three MRI sequences (T2-weighted, AP, and HBP) and fused sequences (combined of three sequences). Nomograms were constructed of a clinical-radiological (CR) model and a clinical-radiological-radiomics (CRR) model. Statistical Tests One-way analysis of variance, independent t-test, Chi-square test or Fisher's exact test, Wilcoxon rank-sum test, LASSO, logistic regression analysis, area under the curve (AUC), nomograms, decision curve, net reclassification improvement (NRI), integrated discrimination improvement (IDI) analyses, and DeLong test. Results Among eight radiomics models, the fused sequences-based VOIintratumor+peritumor radiomics model showed the best performance. The CRR model containing the best performance radiomics model and CR model with the AUC values were 0.934, 0.889, and 0.875, respectively. NRI and IDI analyses showed that the CRR model improved evaluation efficacy over the CR model for all three cohorts (all P-value <0.05). Data Conclusion The CRR model nomogram could preoperatively evaluate MVI in small solitary HCC. The radiomics model based on VOIintratumor+peritumor might achieve better evaluation results. Evidence Level 4 Technical Efficacy Stage 2