Hepatobiliary phase radiomics for identifying subclinical liver dysfunction in compensated chronic liver disease: development and external validation

Purpose: To develop and externally validate a radiomics model derived from gadoxetic acid–enhanced hepatobiliary phase (HBP) MRI for detecting subclinical liver dysfunction in patients with compensated chronic liver disease (CLD). Methods: In this retrospective study, patients were included based on Child–Pugh Class A status and absence of clinical decompensation. Subclinical liver dysfunction was defined using a composite reference standard integrating ALBI score ( > − 2.60), MELD-Na ≥ 10, and FIB-4 ≥ 2.67. Whole-liver radiomic features were extracted from HBP images. Feature selection and model development were performed using Boruta and class-weighted logistic regression within a nested cross-validation framework. Results: A total of 138 individuals (90 patients with subclinical liver dysfunction and 48 controls) were analyzed. Six radiomic features were selected for model development. In the internal test cohort (n = 21), the model achieved a ROC–AUC of 0.86 [95% CI: 0.66–1.00] with an accuracy of 0.81 [95% CI: 0.62–0.95]. In the external validation cohort (n = 26), the model achieved a ROC–AUC of 0.89 [95% CI: 0.73–0.99] and an accuracy of 0.77 [95% CI: 0.58–0.92] using a fixed decision threshold. Both first-order and texture-based features were represented among the selected predictors. Conclusion: HBP-derived radiomics demonstrates potential as a noninvasive biomarker for identifying subclinical liver dysfunction in compensated CLD patients. While the results are promising, the relatively small validation cohorts warrant cautious interpretation, and further validation in larger multicenter populations is required to confirm robustness and generalizability.