Inter-reader agreement of abbreviated magnetic resonance imaging for hepatocellular carcinoma detection: a systematic review and meta-analysis

To determine the inter-reader agreement of abbreviated magnetic resonance imaging (AMRI) for the detection of hepatocellular carcinoma (HCC) and explore the causes of heterogeneity between the reported results.

Original studies reporting the inter-reader agreement of AMRI for detecting HCC were identified in MEDLINE, EMBASE, and Cochrane databases. The pooled kappa coefficient (κ) was calculated using the DerSimonian-Laird random-effects model. Subgroup analyses were performed according to the AMRI protocol (non-contrast [NC]-AMRI, dynamic contrast-enhanced [DCE]-AMRI, and hepatobiliary phase [HBP]-AMRI). Meta-regression analyses were performed to further explore study heterogeneity.

In the eight included studies (1182 patients), the overall pooled κ was 0.76 (95% confidence interval [CI], 0.70-0.82; I² = 74.4%). The κ of NC-AMRI, DCE-AMRI, and HBP-AMRI were 0.72 (95% CI, 0.62-0.82), 0.80 (95% CI, 0.78-0.82), and 0.98 (95% CI, 0.95-1.00), respectively. In the NC-AMRI, the pooled κ of NC-AMRI using only diffusion-weighted imaging (DWI) was 0.64, which was lower than the values using two or more imaging sequences (κ = 0.74-0.77). In subgroup analysis, no study heterogeneity was noted in studies using DCE-AMRI (I² = 0%), whereas high heterogeneity was noted with NC-AMRI (I² = 80.5%). Especially, NC-AMRI including more than two imaging sequences showed high residual heterogeneity (I² = 87.6%). Meta-regression analysis found that difference in reader experience was significantly associated with study heterogeneity (p = .02).

AMRI for detecting HCC showed substantial inter-reader agreement across all examined protocols. NC-AMRI, notably NC-AMRI using only DWI, had relatively low inter-reader agreement. Therefore, DCE-AMRI or HBP-AMRI may be more reliable than NC-AMRI using only DWI.