Inter-reader agreement and imaging-pathology correlation of the LI-RADS M on gadoxetic acid-enhanced magnetic resonance imaging: efforts to improve diagnostic performance.

Abstract

To assess image features for diagnosing LR-M in the Liver Imaging Reporting and Data System (LI-RADS) version 2018 on gadoxetic acid-enhanced magnetic resonance imaging (MRI) and evaluate inter-reader agreement and imaging-pathology correlation. A total of 65 patients diagnosed as LR-M who underwent preoperative gadoxetic acid-enhanced MRI between September 2016 and March 2019 were included retrospectively. All patients were pathologically confirmed and MR images were reviewed by two radiologists without any information of patient history. The intraclass correlation coefficient (ICC) and the Chi-square test were used to evaluate inter-reader agreement and imaging-pathology correlation, respectively. Hepatocellular carcinoma (HCC, n = 23), intrahepatic cholangiocarcinoma (CCA, n = 18), metastasis (n = 10), chronic inflammation (n = 7), combined HCC-CCA (n = 5), sarcoma (n = 1), and neuroendocrine tumor (n = 1) were confirmed. Overall ICC values showed near-perfect to substantial agreement (ranges 0.690-0.887). An assessment of a potential imaging-pathology relationship revealed that HCC, CCA, and chronic inflammation were correlated with 'not showing delayed central enhancement' (p = 0.01), 'other feature suggesting non-HCC malignancy (biliary dilatation and liver surface retraction)' (p = 0.03), and 'infiltrative appearance' (p = 0.00). One or more LR-M image features observed together with biliary dilatation or liver surface retraction, suggested CCA as high specificity (89.4%) and diagnostic accuracy (83.1%). Most of the LR-M image features were not correlated with pathologic diagnosis. However, if one or more of the LR-M features were combined with other well-known image features, diagnostic performance of the LR-M could be improved.