PDF
Citation
Print
Imaging plays a pivotal role in the non-invasive diagnosis of hepatocellular carcinoma (HCC) and in distinguishing it from other liver malignancies, such as intrahepatic cholangiocarcinoma and combined hepatocellular-cholangiocarcinoma. This distinction is essential due to the differing treatment approaches and prognoses for these conditions.1-3 The Liver Imaging Reporting and Data System (LI-RADS), updated in 2018, standardizes liver imaging practices in patients with high-risk HCC.4 It introduces the LR-M category within its diagnostic algorithm to classify lesions that are probably or definitely malignant but not specific for HCC, thereby improving the specificity of HCC diagnoses while retaining sensitivity for detecting general malignancies. The LR-M criteria have evolved to include both targetoid imaging features (such as rim arterial phase hyperenhancement, peripheral washout appearance, delayed central enhancement, targetoid restriction, and targetoid transitional or hepatobiliary phase appearance) and nontargetoid features (such as infiltrative appearance, marked diffusion restriction, and necrosis or severe ischemia), refining the framework for better diagnostic accuracy and reporting consistency in liver imaging.4 Along with the variation in the frequency of LR-M imaging features among non-HCC malignancies,5,6 concerns regarding the significant inter-reader variability for individual LR-M imaging features persist.
In this issue of Journal of Liver Cancer, Kim and Choi7 reported a meta-analytic pooled inter-reader agreement for assessing LI-RADS LR-M features on computed tomography (CT)/magnetic resonance imaging (MRI) and explored the factors contributing to the variability in LR-M assessment. A systematic review and meta-analysis of 24 studies with 5,163 hepatic observations revealed varied pooled κ-values, indicating levels of agreement. Substantial agreement was observed for rim arterial phase hyperenhancement (κ=0.72) and targetoid transitional phase/hepatobiliary phase appearance (κ=0.74), while peripheral washout appearance (κ=0.52) and marked diffusion restriction (κ=0.49) exhibited modest agreement. These results are comparable with inter-reader agreement for assessing LI-RADS major features on MRI, with pooled κ-values ranging from 0.66-0.72.8
In the subgroup meta-regression analysis,7 factors affecting the inter-reader agreement for LR-M imaging features included small observation sizes, the use of 1.5-T MRI scanners, and multiple imaging readers. However, the proportion of patients with liver cirrhosis or non-HCC malignancy, the dominant etiology of liver disease (hepatitis B virus vs. others), and MRI contrast agent (hepatobiliary vs. others) were not significant factors for inter-reader variability among the studies. While these factors account for some variability, they do not provide a complete explanation. Further rigorous analysis is required to identify plausible reasons for inter-reader variability at the individual study level. Factors such as reviewer experience levels, access to previous reports, and study design (prospective vs. retrospective) were not included in the meta-regression analysis due to insufficient information or a lack of prospective studies. Moreover, the inherent subjectivity of qualitative imaging features and varying working and research environments may contribute to unavoidable variability in imaging analysis and LI-RADS categorization. To reduce the subjectivity of qualitative LI-RADS imaging features, quantitative approaches such as radiomics9 can be implemented to improve inter-observer agreement. However, no reliable methodology has yet been established to differentiate between LR-M and other LI-RADS categories.
This systematic review and meta-analysis synthesize results from previous LI-RADS studies on inter-reader agreement, providing significant strengths through its comprehensive approach. However, the interpretation should be approached with caution due to the inherent limitations of the original studies. Most primary studies have been conducted in Northeast Asia (China and South Korea), where hepatitis B is the predominant cause of chronic liver disease. Almost all the included studies utilized MRI, with only one study presenting results using CT. Therefore, it remains unclear whether these results can be generalized to Western populations or if there are differences in interreader agreements across imaging modalities.
LI-RADS aims to enhance patient care through precise liver imaging evaluation. Achieving this goal requires improved interreader agreement on LR-M imaging features. Kim and Choi’s systematic review and meta-analysis7 contribute valuable insights into inter-reader agreement for LR-M imaging features and the sources of variability. Further large-scale multicenter studies on liver imaging and inter-reader agreement in diverse clinical settings are necessary. Evidence-based improvements in LI-RADS diagnostic algorithms and comprehensive educational programs are expected to help meet the objectives of LI-RADS across various clinical practice environments.
XML Download