ML models predict future HCC sites on screening MRI in cirrhosis

Random forest model on screening MRI radiomics predicted future HCC sites in cirrhosis (AUC 0.80, 95% CI 0.78–0.81); AUC 0.86 in LR-3 precursor areas. Heatmaps identified the HCC location in 30% of patients.

• Retrospective study of 121 cirrhotic patients who underwent screening MRI prior to HCC diagnosis.
• Logistic regression, random forest, and XGBoost models compared; best performance with manual VOI annotations.
• In regions with LR-3 precursor lesions, RF AUC rose to 0.86 vs. 0.66 without visible precursor.

Retrospective, single-center design without external validation.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.