A Multicenter Multinational Retrospective Study of the 1-Year Natural History of LI-RADS 3 Observations in Patients with Cirrhosis Luigi Asmundo, Nathaniel Mercaldo, Felipe Furtado et al. Eur Radiol (2026). Published 06 January 2026. https://doi.org/10.1007/s00330-025-12295-5
Hepatocellular carcinoma (HCC) is a leading cause of cancer-related mortality worldwide, primarily affecting patients with cirrhosis. The Liver Imaging Reporting and Data System (LI-RADS) provides standardized categorization for liver observations, where LI-RADS 3 represents "indeterminate" nodules. These observations carry an estimated 15% to 25% risk of progressing to HCC within a year. However, existing evidence has been limited by small, single-center studies. This study aimed to assess the one-year natural history of LI-RADS 3 observations using a large, multinational MRI cohort to identify clinical and imaging predictors of progression. This retrospective study involved six centers across the United States, Italy, and Austria. 347 cirrhotic patients with 540 LI-RADS 3 observations on MRI between January 2023 and March 2024 were analyzed. 78% of patients underwent MRI with extracellular contrast agents, and 22% underwent MRI with hepatobiliary contrast agents. To ensure methodological consistency, only patients with a follow-up MRI at 12 months (±3 months) were included. Exclusion criteria were history of non-HCC malignancy or of HCC, cirrhosis due to congential hepatic fibrosis or vascular disorders, liver transplant after baseline MRI, locoregional therapy before follow-up MRI, insufficient image quality. Experienced radiologists reassessed all observations at baseline and follow-up using LI-RADS v2018 criteria. In addition, the presence and number of co-existing LI-RADS 4 or 5 observations was noted. Clinical data including etiology of liver disease, Child-Pugh class, HBV/HCV status, and liver function tests was collected. To identify predictors of progression (defined as an increase in LI-RADS category), generalized linear mixed-effects (GLME) models and machine learning techniques, including LASSO and random forest models were used. At the one-year mark, 28% (150/540) of LI-RADS 3 observations had progressed to higher categories: 14% (74/540) progressed to LI-RADS 4, 14% (76/540) progressed to LI-RADS 5, and 1% (8/540) progressed to LI-RADS M or TIV. Conversely, 64% remained stable, and 7% regressed to lower categories. Multivariable analysis identified three key independent predictors of progression: Lesion Size: Each additional millimeter in diameter increased the odds of progression (OR: 1.12, 95% CI: 1.01–1.24). Liver Function: Patients with Child-Pugh Class C faced significantly higher risks (OR: 8.36, 95% CI: 1.01–69.27) compared to Class A. Etiology: Alcohol-related liver disease showed an inverse association with progression (OR: 0.24, 95% CI: 0.06–0.94). In univariable analysis, the presence of an enhancing capsule (OR: 23.77), coexistent LI-RADS 5 lesions (OR: 5.38), and untreated HCV status (OR: 2.90) were also significant, though they did not retain independence in the final multivariable model. Notably, ROC analysis identified an optimal lesion size cut-off of 9.5 mm for predicting malignant transformation. The accuracy of the model for predicting progression of LI-RADS 3 lesions significantly improved when combining clinical variables and imaging findings. A model using clinical variables alone (age, sex, etiology, etc.) achieved an AUC of 0.65. Adding imaging features (size, enhancement, washout, capsule) increased the AUC to 0.72 (p=0.01). The study represents the largest cohort to date assessing the 1-year history of LI-RADS 3 observations. The 28% progression rate reinforces the need for close surveillance. The higher risk associated with Child-Pugh Class C likely reflects the interplay between chronic hepatic injury and oncogenesis. Interestingly, the lower progression rate in alcohol-related cirrhosis may be due to a higher prevalence of stable regenerative or hyperplastic nodules that mimic LI-RADS 3 features. Limitations include the retrospective design, which may introduce selection bias. There was also heterogeneity in MRI protocols and contrast agents (hepatobiliary vs. extracellular) across sites, though this was controlled for in the statistical models. Furthermore, the exclusion of patients with history of HCC may have led to an underestimation of the progression rate of LI-RADS 3 lesions. Finally, the study relied on imaging-based criteria rather than histopathological confirmation for all lesions. In conclusion, approximately one in four LI-RADS 3 observations progress within a year. Integrating imaging features—specifically lesion size—with clinical parameters like liver function (Child-Pugh Class) and disease etiology significantly improves risk stratification. These findings support more personalized follow-up strategies and earlier interventions for high-risk indeterminate liver lesions. |
|