Magnetic resonance imaging-based machine learning model for non-invasive liver fibrosis staging: added value to ultrasound liver stiffness measurements(PDF)
《中国医学物理学杂志》[ISSN:1005-202X/CN:44-1351/R]
- Issue:
- 2026年第1期
- Page:
- 41-51
- Research Field:
- 医学影像物理
- Publishing date:
Info
- Title:
- Magnetic resonance imaging-based machine learning model for non-invasive liver fibrosis staging: added value to ultrasound liver stiffness measurements
- Author(s):
- SUN Yiyang1; 2; ZHA Junhao3; ZHANG Chengxiu1; 2; WANG Chenglong1; 2; SONG Yang4; WANG Haijie5; XIA Tianyi3; YANG Guang1; 2
- 1. Shanghai Key Laboratory of Magnetic Resonance, School of Physics and Electronic Science, East China Normal University, Shanghai 200062, China 2. Institute of Magnetic Resonance and Molecular Imaging in Medicine, East China Normal University, Shanghai 200062, China 3. Department of Radiology, Zhongda Hospital, Southeast University, Nanjing 210009, China 4. MR Research Collaboration Team, Siemens Healthineers Ltd., Shanghai 200206, China 5. Innovation Research Institute, Neusoft Medical Systems Co., Ltd., Shenyang 110167, China
- Keywords:
- Keywords: liver fibrosis machine learning magnetic resonance ultrasound transient elastography
- PACS:
- R318；R575
- DOI:
- DOI:10.3969/j.issn.1005-202X.2026.01.006
- Abstract:
- Abstract: Objective To develop and validate a machine learning model based on T1-delay magnetic resonance imaging, referred to as the MR model, for non-invasive liver fibrosis staging, and then combine with or compare against liver stiffness measurements (LSM) derived from ultrasound transient elastography, thereby exploring the added value of MR model to LSM. Methods A retrospective single-center study was conducted on 659 liver fibrosis patients who were randomly divided into training and test cohorts. After image preprocessing, features were selected using hierarchical modeling approach. Specifically, features were categorized into subgroups according to their hierarchical structure, and the performance of subgroup-specific models was evaluated on the validation set to identify key features. The MR model was then established to differentiate between significant liver fibrosis (F≥2) and advanced liver fibrosis (F≥3). Additionally, a MR-LSM model was developed by integrating the MR model with LSM. Model performance was assessed using receiver operating characteristic curves, waterfall plots, and decision curve analysis. Results For the classification of F≥2, the MR-LSM model outperformed both MR model (AUC: 0.824 vs 0.791, P=0.003) and LSM (AUC: 0.824 vs 0.694, P=0.005) in the test cohort. For the classification of F≥3, the MR-LSM model achieved a higher AUC than MR model (0.855 vs 0.819, P=0.004) and LSM (0.855 vs 0.731, P<0.001). Conclusion The MR-LSM model demonstrates superior diagnostic performance in liver fibrosis staging as compared with either MR model or LSM, indicating that the MR model provides substantial added value to LSM.
Last Update: 2026-01-27