#118 - Machine Learning–based Prediction of Suspicious Liver Lesions in Hepatocellular Carcinoma Surveillance Using Alpha-fetoprotein and Pivka-ii.
Year: 2026
Type: article
Study design: Observational Study
Abstract: Introduction and Objectives Introduction: In hepatocellular carcinoma surveillance, abdominal ultrasound and serum alpha fetoprotein is recommended. However, the limited accuracy of alpha fetoprotein can delay early detection of suspicious liver lesions, prompting interest in PIVKA-II as an alternative biomarker. Objective To develop and internally validate a machine learning model combining alpha fetoprotein, PIVKA II, and key clinical and demographic features to predict suspicious ultrasound findings during hepatocellular carcinoma surveillance. Materials and Methods Longitudinal cohort of 300 patients (July 2024–April 2026). The outcome was suspicious ultrasound findings; predictors included AFP, PIVKA-II, age, sex, cirrhosis, and etiology. Biomarker performance was assessed using ROC curves, AUC, and optimal cutoffs for sensitivity and specificity. Penalized logistic regression, Random Forest, and XGBoost were compared using discrimination and accuracy metrics. Internal validation used nested cross validation, and calibration was evaluated with calibration plots and the Hosmer–Lemeshow test. Analyses were performed in R 4.3.0 and Python 3.10 (=0.05). Results Seventeen patients (5.7%) had suspicious ultrasound findings and showed markedly higher AFP and PIVKA II levels (both p<0.001). PIVKA II outperformed AFP, and their combination increased sensitivity (78.4%; AUC 0.864). XGBoost achieved the best performance (AUC 0.947; MCC 0.721). SHAP analysis identified PIVKA II as the most influential predictor, followed by AFP. The high risk group (11% of patients) concentrated over half of positive ultrasound findings. Conclusions Combining AFP and PIVKA II with clinical variables through machine learning produced a highly accurate and clinically useful model for predicting suspicious ultrasound findings, supporting its potential role in personalized surveillance strategies.
Source: Annals of Hepatology
Authors: Dayron Páez, Enrique Arús, Anayda Fernández, Mayra Más, Jorge Hernández, Rolando Castillo, Kenia Valenzuela, Marcia Samada, Mirta Infante, Yusimik Román, Ana Arias, Pedro González, Javier Méndez, Marisley Marrero, Magalys Echemendía, Meilis Mesa, Damaris Herrera, Julio Hernández, Gabriela Hernández, Antonio Díaz
Institutions: Hermanos Ameijeiras Hospital, Instituto de Gastroenterología
Language: English
Subfield: Hepatology
Field: Medicine
Domain: Health Sciences
Open Access status: gold
Estimated APC paid: $1,500
OpenAlex ID: W7204245638