A biology-informed radiomics model for prognostication of hepatocellular carcinoma based on AKR1B10 expression

Background Current radiomic models for hepatocellular carcinoma (HCC) prognosis rely on direct correlations between imaging features and clinical outcomes, resulting in limited biological interpretability and restricted clinical applicability. This study explores a novel biology-driven radiomic strategy focusing on AKR1B10. AKR1B10 is a functionally established molecular driver of HCC progression, and the study aims to develop an interpretable prediction model bridging imaging phenotypes and underlying tumor biology. Methods We analyzed multi-institutional data from The Cancer Genome Atlas (TCGA) and The Cancer Imaging Archive (TCIA). After confirming the prognostic significance of AKR1B10 expression through survival and enrichment analyses, we developed a radiomics model using a cohort of 34 patients with matched computed tomography (CT) images and genomic data. Tumor and peritumoral regions were segmented, and 107 radiomic features were extracted. Feature selection was performed using maximum-relevance-minimum-redundancy (mRMR) and recursive feature elimination (RFE) algorithms, with subsequent model building via logistic regression. The model was evaluated using ROC analysis, calibration curves, and decision curve analysis. Finally, we constructed a prognostic nomogram integrating the radiomics signature with clinical variables. Results AKR1B10 overexpression was significantly associated with poor overall survival (HR = 2.187, 95 % CI: 1.385–3.454, P < 0.001) and characteristic activation of oncogenic pathways. The radiomics model demonstrated strong performance in predicting AKR1B10 status (AUC = 0.83, 95 % CI: 0.69–0.97), with significant difference in rad-scores between AKR1B10 high- and low-expression groups (P < 0.001). The integrated nomogram showed excellent predictive accuracy for 3-year survival (AUC = 0.85) and provided clinical net benefit across threshold probabilities. Conclusions The biology-informed radiomics model based on AKR1B10 expression demonstrates strong prognostic performance in hepatocellular carcinoma. By directly linking imaging phenotypes to a key molecular driver of HCC, this approach provides a clinically applicable and biologically interpretable tool for pre-operative risk prediction.