Interobserver Variability and Diagnostic Performance of Gadoxetic Acid-enhanced MRI for Predicting Microvascular Invasion in Hepatocellular Carcinoma.

Abstract

Background Accurate identification of microvascular invasion (MVI) in hepatocellular carcinoma (HCC) before treatment is critical for selecting a proper treatment strategy. Purpose To evaluate the interobserver agreement and the diagnostic performance of the MRI assessment of MVI in HCC according to the level of radiologist experience. Materials and Methods This retrospective study included 100 patients with surgically confirmed HCCs smaller than 5 cm who underwent gadoxetic acid-enhanced MRI between 2013 and 2016. Eight postfellowship radiologists (four with 7-13 years of experience [more experienced] and four with 3-6 years of experience [less experienced]) evaluated four imaging features (nonsmooth tumor margin, irregular rim-like enhancement in the arterial phase, peritumoral arterial phase hyperenhancement, peritumoral hepatobiliary phase hypointensity) and assigned the possibility of MVI. Interobserver agreement was determined by using Fleiss κ statistics according to reviewer experience and tumor size (≤3 cm vs >3 cm). With reference standards of histopathologic specimens, the diagnostic performance in the identification of MVI was assessed by using receiver operating characteristic curve analysis. Results In 100 patients (mean age, 58 years ± 10 [standard deviation]; 70 men) with 100 HCCs (mean size, 2.8 cm ± 0.9), 39 (39%) HCCs had MVI. The overall interobserver agreement was fair to moderate for the imaging features and their combinations (κ = 0.38-0.47) and MVI probability (κ = 0.41; 95% confidence interval: 0.33, 0.45). More experienced reviewers demonstrated higher agreement in MVI probability than less experienced reviewers (κ = 0.55 vs 0.36, respectively; P = .002). Diagnostic performance of each reviewer was modest for MVI prediction (area under the receiver operating characteristic curve [AUC] range, 0.60-0.74). The AUCs for the diagnosis of MVI were lower for HCCs larger than 3 cm (range, 0.55-0.69) than for those less than or equal to 3 cm (range, 0.59-0.75). Conclusion Considerable interobserver variability exists in the assessment of microvascular invasion in hepatocellular carcinoma using MRI, even for more experienced radiologists. © RSNA, 2020 Online supplemental material is available for this article. See also the editorial by Tang in this issue.