Machine Learning Model for Non-invasive Liver Fibrosis Staging on MRI and its Added Value to Ultrasound Liver Stiffness Measurements

Motivation: Early non-invasive assessment of liver fibrosis is crucial for timely intervention and treatment. Goal(s): This study aims to explore the added value of T1-delay MR to liver stiffness measurement (LSM). Approach: We retrospectively collected of 655 chronic hepatitis B patients and constructed machine learning models based on T1-delay MR images and ultrasound LSM for liver fibrosis staging. Results: The MR-LSM model achieved an AUC of 0.824 for identification of clinically significant liver fibrosis (F≥2), outperforming the standalone MR and LSM models. For identification of advanced fibrosis (F≥3), the MR-LSM model reached an AUC of 0.855, significantly surpassing other models. Impact: T1-delay MR can be combined with LSM for non-invasive liver fibrosis classification, enabling early detection of liver fibrosis and timely intervention.