As the hallmark of cancer, genetic and phenotypic heterogeneity leads to biomarkers that are typically tailored to specific cancer type or subtype. This specificity introduces complexities in facilitating streamlined evaluations across diverse cancer types and optimizing therapeutic outcomes. In this study, we comprehensively characterized the radiological patterns underlying liver cancer (LC) by integrating radiomics profiles from computed tomography (CT) images of hepatocellular carcinoma (HCC), intrahepatic cholangiocarcinoma (ICC), and colorectal cancer liver metastases (CRLM) through unsupervised clustering analysis. We identified three distinct radiomics clusters, displaying heterogeneity in prognosis. Subsequently, we formulated a shared prognosticator, the liver cancer radiomics signature (LCRS), by discovering and manifesting connectivity among radiomics phenotypes using GGI strategy. We validated that the LCRS is independent prognostic factor after adjusting for clinic-pathologic variables (all P < 0.05), with the LCRS-High group consistently associated with worse survival outcomes across HCC, ICC, and CRLM. However, the LCRS-High group showed clinical benefit from adjuvant chemotherapy, leading to reduced disease recurrence risk and improved survival. By contrast, the LCRS-Low group, including a subset of gastric cancer liver metastases (GCLM), exhibited more favorable response to immune checkpoint inhibitors (ICIs)-based combinational therapy (P = 0.02, hazard ratio (HR): 0.34 [95 % confidence interval (CI): 0.13–0.88]). Further analysis revealed that Notch signaling pathway was enriched in LCRS-High tumors, while LCRS-Low tumors exhibited higher infiltration of natural killer cell. These findings highlight the promise of this universal scoring model to personalize management strategies for patients with LC.
Read full abstract