AI and Cancer Research ── Liver ── 2026-11-05
Liver cancer: treatment, trials and AI
- Abstract
Objective: Microvascular invasion (MVI) predicts postoperative recurrence in hepatocellular carcinoma (HCC), yet preoperative prediction remains challenging. This study compared a Transformer, a convolutional neural network (CNN), and a multiple instance learning (MIL) framework for predicting MVI from contrast-enhanced CT.
Methods: We retrospectively enrolled 236 patients with pathologically confirmed HCC. Tumor regions of interest were semi-automatically delineated on arterial-phase CT images, and slices centered on the largest tumor cross-section were sampled to build a 2.5D dataset. Four models—Transformer, MIL, CNN, and a clinical baseline—were trained. Models were compared using the area under the receiver operating characteristic curve (AUC) with the DeLong test; bootstrap resampling assessed the stability of AUC estimates.
Results: The Transformer achieved the highest AUC of 0.827 (95% CI 0.7585–0.8962), with an accuracy of 0.775, sensitivity of 0.577, and specificity of 0.889, followed by MIL (0.766; 95% CI 0.6875–0.8441), CNN (0.723), and the clinical model (0.715). Bootstrap validation confirmed stable AUC estimates for all models, with the Transformer ranking first (mean 0.827; SD 0.034). The Transformer significantly outperformed the CNN (adjusted P = 0.001) and the clinical model (adjusted P = 0.039); after correction for multiple comparisons, its difference from MIL was of borderline significance (adjusted P = 0.097). No significant differences were observed among the CNN, MIL, and clinical models.
Conclusion: Among the four models evaluated, the contrast-enhanced CT-based Transformer predicted MVI most accurately, and its advantage persisted after bootstrap internal validation. These findings concern the ranking of discriminative performance only; the clinical utility of the model is not established, and prospective validation in larger cohorts is required before any clinical application.
Journal IF-equivalent: 2.8 (OpenAlex 2-year mean citedness, value as of 2026-11-03, retrieved 2026-11-04; not the official Clarivate IF)Reference: Lin M, Zheng X, Zhao C, Lin T, Zhao Q. Predicting microvascular invasion in hepatocellular carcinoma using contrast-enhanced CT-based deep learning models: a comparative study of transformer, CNN, and MIL. Front Oncol. 2026 Nov 2;16:1988882. doi:10.3389/fonc.2026.1988882.Checked: Abstract only - Abstract
INTRODUCTION: Hepatocellular carcinoma (HCC) exhibits substantial molecular heterogeneity, complicating treatment selection and clinical management. OBJECTIVES: We developed an artificial intelligence-based histopathological framework to predict HCC biological characteristics from routine hematoxylin and eosin (H&E)-stained slides. METHODS: Using The Cancer Genome Atlas Liver Hepatocellular Carcinoma (TCGA-LIHC) cohort, transcriptomic phenotypes associated with angiogenesis, stemness, and autophagy were identified through gene set analysis, differential expression analysis, and consensus clustering, and evaluated for overall survival and drug sensitivity. Whole-slide images were processed into 224 × 224-pixel tumor tiles for phenotype prediction using feature extraction (UNI, ResNet50, CTrans, GigaPath, and Virchow2) combined with classic multiple instance learning and clustering-constrained attention-based multiple instance learning.. Models were externally validated at Renmin Hospital of Wuhan University and Tongji Hospital. Interpretability was assessed using attention heatmaps, and predictions were evaluated against molecular markers and tumor differentiation grades. RESULTS: Unsupervised clustering identified two subgroups per phenotype, with significant difference in survival (all p < 0.004) and sensitivities to sorafenib or transcatheter arterial chemoembolization (all p < 0.0001). The tumor detection model achieved an area under the receiver operating characteristic curve (AUC) of 97.53%. The phenotype prediction models achieved internal optimal AUCs of 85.51%-91.05% and accuracies of 79.63%-88.89%. External optimal AUCs were 70.94%, 88.64%, and 91.17% for angiogenesis, stemness, and autophagy, with accuracies of 62.00%, 78.00%, and 84.00%, respectively. Attention heatmaps revealed distinct morphological features among subgroups, including neovascularization, increased nuclear-to-cytoplasmic ratios, and cytoplasmic vacuolization. Model predictions showed concordance with CD31/CD105 (κ = 0.398-0.432), CD133/LGR5 (κ = 0.428-0.629), Beclin-1/p62-LC3B (κ = 0.592-0.762) (all p < 0.05), and tumor differentiation grade (κ = - 0.167 to - 0.175, p < 0.05). CONCLUSION: This proof-of-concept AI framework predicted transcriptome-defined HCC phenotypes from routine histology, with predictions associated with prognosis and therapeutic response. It may provide a cost-effective approach for assessing HCC heterogeneity and complement molecular profiling.
Journal IF-equivalent: 9.9 (OpenAlex 2-year mean citedness, value as of 2026-11-03, retrieved 2026-11-04; not the official Clarivate IF)Reference: Zhang H, Li J, Sheng J, Li Z, Jiang W, Xu K, et al. Rapid and label-free analysis of biological characteristics of hepatocelluar carcinoma tissues based on histological images. J Adv Res. 2026 Nov [Epub ahead of print]. doi:10.1016/j.jare.2026.10.067. PMID: 42967526.Checked: Abstract only