Token-guided multimodal prognosis in hepatocellular carcinoma: a framework steered by tumour-stroma ratio - PubMed
5 hours ago
- #Artificial Intelligence
- #Tumor-Stroma Ratio
- #Hepatocellular Carcinoma
- The tumour-stroma ratio (TSR) is a potential prognostic indicator in hepatocellular carcinoma (HCC), but its quantification is challenging.
- The study aimed to determine if TSR follows a non-linear prognostic pattern and to develop an AI-powered framework for standardized TSR assessment and prognosis prediction.
- Researchers integrated whole-slide image (WSI) data with clinical variables from a retrospective cohort (n=392) and The Cancer Genome Atlas dataset (n=168).
- An inverted U-shaped non-linear relationship between TSR and mortality was identified, with a risk initiation threshold at 0.188 and a peak at 0.268.
- Transcriptomics analysis revealed that the high-risk phenotype is characterized by active tumor proliferation, stromal activation, and tumor microenvironment crosstalk.
- AI-derived TSR showed strong correlation with expert assessment (R² >0.9).
- A novel 'Token-Guided Multimodal Fusion' architecture was developed to integrate WSI, TSR, and clinical variables as high-dimensional tokens.
- The multimodal framework demonstrated high prognostic accuracy (AUC >0.80) compared to unimodal baselines.
- The study suggests that the future of computational pathology lies in the semantic fusion of human domain knowledge with AI reasoning, rather than simple quantification.