Value of radiomic analysis of data from magnetic resonance elastography for diagnosing fibrosis stages in patients with hepatitis B/C

Abstract

To evaluate the value of radiomic analysis of data from MR elastography (MRE) for diagnosing stages of fibrosis in patients with chronic hepatitis B/C (CHB/CHC). MRE images from 140 CHB/CHC patients with pathologically proven F0–F4 hepatic fibrosis stages were retrospectively analyzed, which is approved by our Institutional Review Board (IRB). A total of 364 radiomic features were extracted from spin-echo echo planar imaging with three-dimensional postprocessing (EPI3D) and gradient recalled echo with two-dimensional postprocessing (GRE2D) MRE data, including histogram parameters (HPs), gray level co-occurrence matrix (GLCM), and gray level run length matrix. Feature selection (significance test and correlation analysis) and classification (logistic regression models) were performed to stage ≥ F1, ≥ F2, ≥ F3 and F4, with a leave-one-out-based cross-validation method. Receiver operating characteristic analysis including areas under the curves (AUCs) was performed. Univariate analysis showed the best-performing features for classifying fibrosis stages were HPs (AUC range 0.798–0.929 for EPI3D and 0.727–0.916 for GRE2D). In multivariate radiomic analysis, AUCs for diagnosing fibrosis of ≥ F1, ≥ F2, ≥ F3 and cirrhosis (F4) were 0.888, 0.934, 0.924 and 0.948 for EPI3D MRE, and 0.887, 0.867, 0.892 and 0.941 for GRE2D MRE, which were slightly higher than that of EPI and GRE stiffness for each fibrosis stage (EPI 0.816, 0.924, 0.870 and 0.918; GRE 0.882, 0.800, 0.865 and 0.911). The AUCs of the EPI3D model were significantly higher than GRE2D model for staging clinically significant fibrosis (0.934 vs 0.867, P = 0.030) and advanced fibrosis (0.924 vs 0.892, P = 0.029). Radiomic analysis of MRE facilitates fibrosis classification in CHB/CHC patients, especially for identifying intermediate fibrosis, with a higher diagnostic performance by the EPI3D MRE.