Radiomic features disclose the presence of microvascular invasion in hepatocellular carcinoma

Purpose or Learning Objective To investigate whether a radiomic analysis performed on hepatocellular carcinoma can reveal the presence of microvascular invasion (MVI). Methods or Background The dataset includes seventy-one early-stage HCC nodules (Milan criteria, nodule≤3 cm, without macrovascular invasion and extrahepatic spread), where forty-two are MVI positive (MVI+) and twenty-nine are MVI negative (MVI-), at histopathological report obtained after surgery. Radiomic analysis is performed on contrast-enhanced CT imaging during the arterial phase. HCC nodules were manually segmented and eighty-four local first-order radiomic features (RFs) are computed from HCC Region of Interest (ROIs). To avoid overfitting, only one couple of RFs is selected for discriminating MVI+ and MVI- into three steps: (i) a subset of RFs is first selected through LASSO; (ii) linearly correlated couples are discarded; (iii) after computing the ROC curve, the most discriminating couple is selected as that one yielding the highest AUC. The discrimination between MVI+ and MVI- is assessed through specificity and sensitivity computed at the Youden Index (YI). Results or Findings The selected couple combines a measure of local heterogeneity to mean CT image values in HCC ROIs and allows achieving an AUC=0.86, with specificity=81% and sensitivity=83% at the YI=0.64. Conclusion The radiomic analysis allows unravelling different inner properties of HCC MVI+ nodules, which show a more heterogeneous tumour pattern with respect to MVI-. Limitations This radiomic analysis was performed exclusively on arterial-phase contrast enhanced images, although it is reasonable arguing that the next inclusion of venous phase images must only improve the outcomes.