NGHIÊN CỨU GIÁ TRỊ CÁC MÔ HÌNH HỌC MÁY TRONG CHẨN ĐOÁN XƠ HÓA GAN ĐÁNG KỂ TRÊN CƠ SỞ DỮ LIỆU SIÊU ÂM 2D VÀ XÉT NGHIỆM SINH HÓA GAN

Abstract

Objectives: To evaluate the value of machine learning models in diagnosing significant liver fibrosis (≥ F2) based on routine ultrasound and liver biochemical data. Methods: A cross-sectional descriptive study was conducted on 138 patients with liver disease (hepatitis B, hepatitis C, alcohol-related liver disease, and nonalcoholic fatty liver disease) from January to December 2020. Patients were assessed using biochemical tests (AST, ALT, GGT), 2D ultrasound, and Fibroscan to classify liver fibrosis severity (F0 - F1 vs. F2 - F4). Three machine learning models (decision tree, random forest, and gradient boosting) were developed to classify ≥ F2 fibrosis using blood test and 2D ultrasound data, with 80% of the data for training and 20% for testing. Results: In the training dataset, the diagnostic performance of the machine learning models yielded area under the curve (AUC) values of 0.85, 0.93, and 0.96 for decision tree, random forest, and gradient boosting, respectively. In the testing dataset, AUC values were 0.84, 0.88, and 0.89, respectively. Conclusion: Machine learning models demonstrated high diagnostic performance for significant liver fibrosis, supporting physicians in non-invasively assessing the degree of liver fibrosis in patients with liver disease.