Clustering of quantitative CT features identifies HCC subtypes with distinct prognosis and immune signatures

OBJECTIVE: There is still a lack of widely applicable biomarkers for immunotherapy in hepatocellular carcinoma (HCC). We aim to identify subtypes using computed tomography (CT) imaging features of HCC and assess their value in predicting prognosis and the effectiveness of immunotherapy. MATERIALS AND METHODS: We used unsupervised consensus clustering based on quantitative contrast-enhanced CT features to identify imaging subtypes and investigated their value in predicting prognosis and the effectiveness of immunotherapy in the discovery (n = 103) and immunotherapy-treated (n = 110) validation cohort. We also developed a gene-based classifier for imaging subtypes and tested their prognostic and biological relevance in two additional gene validation cohorts with publicly available gene expression data but without imaging data (n = 551). RESULTS: The imaging subtypes demonstrated significant correlations with overall survival across all cohorts (p = 0.002, p < 0.001) and effectively predicted immunotherapy outcomes, with 1-year progression-free survival rates varying significantly (p < 0.001). Imaging subtypes 1 and 3 with favorable and intermediate immunotherapy responses showed significant activation of B-cells, lymphocyte-mediated immunity, and immune response (adjusted p-value ≤ 0.047, false discovery rate ≤ 0.043). Imaging subtype 2 with poor immunotherapy responses displayed the least activated CD8+ T cells, the lowest cytolytic activity, and the highest infiltration of neutrophil and regulatory T cells. Additionally, type I and II interferon responses were significantly downregulated in imaging subtype 2. CONCLUSION: Unsupervised clustering of CT imaging features identified subtypes with significantly distinct prognosis and immune signatures. The imaging subtypes can serve as a marker for immunotherapy in HCC. RELEVANCE STATEMENT: The imaging subtypes identified in this study demonstrate significant clinical relevance by providing a non-invasive approach to predict prognosis and immunotherapy response for HCC. KEY POINTS: We identified three novel imaging subtypes using unsupervised consensus clustering of quantitative CT imaging features for HCC. Imaging subtypes were independent predictors of overall survival and the effectiveness of immunotherapy for HCC. The imaging subtypes can be used as a marker for immunotherapy in HCC.