Machine Learning Stratification of Liver Stiffness using T2-weighted MRI Radiomic Data: A Multi-Site Study

MR elastography (MRE) offers a non-invasive approach to quantify liver stiffening, a surrogate for hepatic fibrosis. However, it has drawbacks, including long exam time, patient discomfort, and the need for additional hardware. The objective of this multi-site study is to develop a machine learning model to categorically stratify the severity of liver stiffness using clinical, routinely collected T2-weighted MRI data from pediatric and adult patients from four study sites. With radiomic features extracted from MRI data, our model achieved an AUROC of 0.72 for stratifying liver stiffness, demonstrating the potential of such a machine learning strategy for clinical utilization.