AI and Cancer Research ── Liver ── 2026-10-28
Liver cancer: treatment, trials and AI
- Abstract
Vessels encapsulating tumor clusters (VETC) are significantly associated with poor prognosis in hepatocellular carcinoma (HCC), with implications for patient survival and clinical management. This study aimed to evaluate a deep learning (DL) radiomics model based on Gd-EOB-DTPA-enhanced MRI for preoperative prediction of the VETC pattern in HCC. Patients with pathologically confirmed HCC who underwent Gd-EOB-DTPA-enhanced MRI between January 2016 and August 2024 were retrospectively enrolled. Visually assessed radiological features were collected to construct a clinical-radiological (CR) model. Deep learning and handcrafted radiomics features were extracted from arterial, portal venous, and hepatobiliary phases. Feature selection and DL radiomics model construction were performed using the least absolute shrinkage and selection operator (LASSO). Multivariable logistic regression was applied to develop a hybrid model and corresponding nomogram. Model discrimination was quantified using the area under the receiver operating characteristic curve (AUC). Cox regression analysis was performed to identify predictors of recurrence-free survival (RFS). In the training cohort, the hybrid DL-radiomics nomogram yielded the highest AUC (0.92), compared with 0.87 for the DL model, 0.80 for the radiomics model, and 0.74 for the CR model. However, standalone DL model showed better discriminative performance in the internal validation cohort, with an AUC of 0.87, sensitivity of 0.80, and specificity of 0.87. VETC status, AST level, and intratumoral necrosis were identified as independent predictors of RFS and incorporated into the VAI prognostic model. The VAI model achieved an AUC of 0.75 (95% CI, 0.68-0.80), which was significantly higher than those of three conventional prognostic systems (range, 0.52 [95% CI, 0.46-0.59] to 0.58 [95% CI, 0.51-0.64]). The hybrid DL-radiomics model based on Gd-EOB-DTPA-enhanced MRI showed favorable discriminative performance for VETC prediction in the training cohort, while the standalone DL model demonstrated better performance in the internal validation cohort. The VAI model enables individualized prediction of postoperative RFS for patients undergoing hepatectomy.
Journal IF-equivalent: 3.9 (OpenAlex 2-year mean citedness, value as of 2026-10-13, retrieved 2026-10-15; not the official Clarivate IF)Reference: Wang M, Wang Y, Cao L, Shen Y, Xi D, Yan R, et al. Deep Learning Radiomics to Predict Vessels Encapsulating Tumor Clusters and Prognosis in Hepatocellular Carcinoma Based on Gd-EOB-DTPA-Enhanced MRI. J Digit Imaging Inform med. 2026 Oct 26 [Epub ahead of print]. doi:10.1007/s10278-026-02456-3.Checked: Abstract only