Noninvasive Grading of Liver Fibrosis Based on Texture Analysis From MRI‐Derived Radiomics

ABSTRACT Given the increasing global prevalence of metabolic syndrome, this study aimed to assess the potential of MRI‐derived radiomics in noninvasively grading fibrosis. The study included 79 prospectively enrolled participants who had undergone MRE due to known or suspected liver disease between November 2022 and September 2023. Among them, 48 patients were diagnosed with histopathologically confirmed liver fibrosis. A total of 107 radiomic features per patient were extracted from MRI imaging. The dataset was then divided into training and test sets for model development and validation. Stepwise feature reduction was employed to identify the most relevant features and subsequently used to train a gradient‐boosted tree model. The gradient‐boosted tree model, trained on the training cohort with identified radiomic features to differentiate fibrosis grades, exhibited good performances, achieving AUC values from 0.997 to 0.998. In the independent test cohort of 24 patients, the radiomics model demonstrated AUC values ranging from 0.617 to 0.830, with the highest AUC of 0.830 (95% CI 0.520–0.830) for classifying fibrosis grade 2. Incorporating ADC values did not improve the model's performance. In conclusion, our study emphasizes the significant promise of using radiomics analysis on MRI images for noninvasively staging liver fibrosis. This method provides valuable insights into tissue characteristics and patterns, enabling a retrospective liver fibrosis severity assessment from nondedicated MRI scans. This study demonstrated that an MRI‐based radiomics model using a gradient‐boosted tree could reliably classify liver fibrosis stages, achieving AUC values up to 0.830 in the test cohort. The model performed best in identifying fibrosis grade 2, with no additional benefit from incorporating ADC values. These findings highlight radiomics' potential for noninvasive, retrospective liver fibrosis grading using standard MRI scans.