Texture Analysis of Non-contrast Magnetic Resonance Imaging for Liver Fibrosis Staging in Patients with Chronic...

To evaluate the diagnostic performance of texture analysis parameters obtained from non-contrast liver magnetic resonance imaging (MRI) in staging liver fibrosis in patients with chronic hepatitis B.

Non-contrast MRI examinations of 66 patients with chronic hepatitis B were retrospectively evaluated. The protocol included in-phase T1-weighted (IP-T1W), out-of-phase T1-weighted (OP-T1W), fat-suppressed T1-weighted (FS-T1W), fat-suppressed T2-weighted (FS-T2W), and T2-weighted (T2W) sequences. Histogram-based texture features were extracted from regions of interest placed in the liver parenchyma. For each sequence, the most informative features were selected from 20 candidate variables using the minimal-redundancy maximal-relevance (mRMR) algorithm, and multivariable logistic regression models were constructed. Fibrosis was categorized as early fibrosis (F0-F2) or significant fibrosis (F3-F6). Diagnostic performance was assessed using receiver operating characteristic (ROC) analysis and area under the curve (AUC), and model generalizability was evaluated by 10-fold cross-validation.

Sixty-six patients were included (early fibrosis, n=43; significant fibrosis, n=23). ROC analysis showed that the highest discriminatory performance was obtained with the OP-T1W sequence (AUC=0.789, 95%CI: 0.677-0.900, p<0.001), while the IP-T1W sequence also demonstrated acceptable diagnostic performance (AUC=0.723, 95%CI: 0.600-0.846, p=0.002). The FS-T1W, FS-T2W, and T2W sequences showed lower AUC values of 0.686, 0.674, and 0.657, respectively. DeLong test comparisons revealed no statistically significant differences between the AUC values of the models derived from different MRI sequences (all p>0.05). Based on the Youden index-derived thresholds, the OP-T1W sequence achieved the highest diagnostic accuracy (72.31%) and sensitivity (79.17%), whereas the FS-T1W sequence demonstrated the highest specificity (83.72%). In 10-fold cross-validation, the highest mean AUC was observed for OP-T1W (0.785±0.168), followed by FS-T2W (0.760±0.168), whereas IP-T1W showed a lower mean AUC (0.653±0.128).

Texture analysis parameters derived from routine non-contrast MRI sequences in patients with chronic hepatitis B may provide additional quantitative information for the noninvasive assessment of liver fibrosis. These findings suggest that texture analysis may serve as a complementary tool within routine liver MRI. However, given the single-center design and limited sample size of the present study, further studies with larger cohorts and external validation are needed before clinical application.