Effect of combining serum alpha-fetoprotein with LI-RADS v2018 on gadoxetate-enhanced MRI in the diagnosis and prognostication of hepatocellular carcinoma

Objectives To evaluate the performance of combining serum alpha-fetoprotein (AFP) with LI-RADS (LI-RADS + AFP) on gadoxetate-enhanced MRI for diagnosing hepatocellular carcinoma (HCC) <= 3.0 cm and predicting post-surgical patient outcomes, compared to LI-RADS v2018. Methods Patients with hepatic observations <= 3.0 cm who underwent preoperative gadoxetate-enhanced MRI and surgical resection were retrospectively analyzed. In LI-RADS + AFP, LR-4 (probably-HCC), LR-M (probably or definitely-malignant but not HCC specific), or LR-TIV (definite-tumor-in-vein) by LI-RADS v2018 was upgraded to LR-5 (definitely-HCC) if they met AFP criteria based on three different cut-offs (>= 20, >= 200, >= 400 ng/mL). The sensitivity, specificity, and accuracy of LI-RADS v2018 LR-5 and LI-RADS + AFP LR-5 were compared using generalized estimating equations. Recurrence-free survival (RFS) and overall survival (OS) in the two groups (LR-4/5 vs. LR-M) of LI-RADS v2018 and LI-RADS + AFP were compared using log-rank tests. Results In 520 observations from 414 patients (mean age, 60 +/- 9 years; 328 men), LI-RADS + AFP >= 20 ng/mL (LI-RADS + AFP20) showed the highest accuracy among the three cut-offs. Compared to LI-RADS v2018, LI-RADS + AFP20 showed significantly higher sensitivity in diagnosing HCC (79.9% [95% confidence interval, 75.7-83.6] vs. 71.8% [67.2-76.1], p < 0.001) and accuracy (82.7% [79.2-85.9] vs. 76.9% [73.1-80.5], p < 0.001) without difference in specificity (93.5% [87.1-97.4] vs. 96.3% [90.8-99.0], p = 0.083). Unlike LI-RADS v2018, showing no difference in RFS between LR-4/5 and LR-M (50.6 vs. 42.6 months; p = 0.078), LI-RADS + AFP20 showed significant differences in RFS (50.6 vs. 35.0 months; p < 0.001) and OS (67.6 vs. 54.8 months, p = 0.008). Conclusion Combining AFP and LI-RADS v2018 resulted in better diagnostic performance and superior post-surgical prognostication compared to LI-RADS v2018 alone.