To investigate the impact of preoperative gadoxetate disodium (EOB) MRI-based radiomics on predicting glypican 3 (GPC3)-positive expression and the relevant recurrence-free survival (RFS) of HCC ≤ 5 cm.Between January 2014 and October 2018, 259 patients with solitary HCC ≤ 5 cm who underwent hepatectomy and preoperative EOB-MRI were retrieved. Multivariate logistic regression was implemented to identify independent predictors for GPC3. By combining five feature selection strategies and three classifiers, 15 GPC3-oriented radiomics models could be constructed, the best of which with independent clinicoradiologic predictors was integrated into the comprehensive nomogram.GPC3 was an independent risk factor of postoperative recrudescence for HCC. Alpha-fetoprotein20 ng/mL, homogenous T2 signal and hypointensity on hepatobiliary phase were independently related to GPC3-positive expression in the clinicoradiologic model. With 10 features selected by support vector machines-recursive feature elimination, logistic regression-based classifier achieved the best performance among 15 radiomics models. After five-fold cross-validation, our comprehensive nomogram acquired better average area under receiver operating characteristic curves (training and validation cohorts: 0.931 vs. 0.943) than the clinicoradiologic algorithm (0.738 vs. 0.739) and the optimal radiomics model (0.943 vs. 0.931). Net reclassification indexes further demonstrated the superiority of GPC3 nomogram over clinicoradiologic and radiomics algorithms (46.54%, p0.001; 7.84%, p = 0.207). Meanwhile, higher radiomics score significantly shortened the median RFS (from77.9 to 48.2 months, p = 0.044), which was analogue to that of the histological GPC3-positive phenotype (from73.9 to 43.2 months, p0.001).Preoperative EOB-MRI radiomics-based nomogram satisfactorily distinguished GPC3 status and outcomes of solitary HCC ≤ 5 cm.
Full Text via ClinicalKey