Interreader Agreement for Individual Ancillary Features in LI-RADS CT/MRI Version 2018: A Systematic Review and Meta-Analysis.

BACKGROUND. In LI-RADS CT/MRI version 2018 (v2018), ancillary features (AFs) support observation characterization and category assignment. The literature is lacking regarding interreader agreement on a broad range of AFs. OBJECTIVE. The purpose of the present study was to conduct a systematic review and meta-analysis of the interreader agreement for all individual AFs in LI-RADS CT/MRI v2018 and to identify factors associated with interpretation variability. EVIDENCE ACQUISITION. The MEDLINE, Embase, and Cochrane databases were searched for original research studies published from January 1, 2018, to September 15, 2025, that reported interreader agreement for individual AFs in LI-RADS CT/MRI v2018. Pooled kappa values for the 21 individual AFs were determined using a DerSimonian-Laird random-effects model with Knapp-Hartung adjustment. Study characteristics independently associated with agreement were identified by metaregression analyses. EVIDENCE SYNTHESIS. The final analysis included 47 studies comprising a total of 11,891 patients and 13,695 hepatic observations. Three AFs (ultrasound visibility as discrete nodule, size stability for ≥ 2 years, and size reduction) were not reported in any studies. The pooled kappa value among AFs favoring malignancy but not HCC in particular was 0.58 for subthreshold growth, 0.63 for corona enhancement, 0.69 for fat sparing in solid mass, 0.70 for transitional phase hypointensity, 0.72 for iron sparing in solid mass, 0.75 for restricted diffusion, 0.77 for mild-to-moderate T2 hyperintensity, and 0.84 for hepatobiliary phase (HBP) hypointensity; among AFs favoring HCC in particular, it was 0.58 for nonenhancing capsule, 0.66 for nodule-in-nodule appearance, 0.69 for mosaic architecture, 0.72 for fat in mass more than adjacent liver, and 0.75 for blood products in mass; and among AFs favoring benignity, it was 0.26 for enhancement paralleling blood pool, 0.58 for HBP isointensity, 0.59 for marked T2 hyperintensity, 0.60 for undistorted vessels, and 0.73 for iron in mass more than liver. In metaregression analyses, study characteristics showing significant independent associations with greater agreement for at least one AF included a proportion of patients with cirrhosis that was less than 80% and modality use of MRI only. CONCLUSION. Interreader agreement was generally higher for AFs favoring malignancy than for those favoring benignity. CLINICAL IMPACT. Cautious application of AFs in clinical practice is advised for the AFs associated with low agreement, particularly in settings further contributing to reduced agreement.