Cross-modality fusion of hematology-based digital and molecular biomarkers for intelligent epidemiological screening and surveillance of hepatocellular carcinoma.

Early detection of hepatocellular carcinoma (HCC) remains a critical challenge due to the disease's complexity and the limitations of current screening and surveillance methods, which are often invasive, costly, or lack adequate sensitivity. There is an increasing demand for accessible methods with explicit and reliable indicators capable of detecting HCC early, ideally before symptoms manifest. Here, we propose an intelligent, hematology-based testing principle and a platform that integrates cross-modality digital and serum biomarkers to provide a comprehensive characterization of multi-faced HCC progression. To enhance the sensitivity, we developed a surface-enhanced Raman spectroscopy (SERS) biosensor with an enhancement factor above 108. Leveraging explainable artificial intelligence, we identified hematology-based digital biomarkers, a compendium of the 15 most salient features per class (normal, high-risk, and HCC) extracted from the hematological SERS spectra generated by the biosensor. A methodological analysis of the cross-modality biomarker fusion strategies suggests that decision-level fusion yields the best performance. When combined with clinical data, the resulting model substantially improved accuracy to 0.99 for early-stage HCC detection, surpassing existing models like GALAD, which achieved an accuracy of 0.83. This study demonstrates a way to deriving actionable reference for liquid biopsy with the potential to advance early cancer detection.