Hepatocellular carcinoma has a poor prognosis, and response to immune checkpoint inhibitors remains limited. Better markers are needed to identify patients more likely to benefit from treatment. A recent analysis published in European Radiology Experimental examined whether computed tomography features can identify hepatocellular carcinoma subtypes linked to prognosis and immunotherapy response. Computed tomography may offer a non-invasive way to assess tumour biology, especially when diagnosis can rely on imaging alone.
CT Features Identify Three Tumour Groups
The framework uses contrast-enhanced computed tomography scans from arterial and portal venous phases. Quantitative imaging features are extracted from hepatocellular carcinoma lesions after tumour segmentation, image processing and standardisation. Unsupervised consensus clustering then groups tumours according to imaging patterns rather than established clinical categories.
Must Read: Post-Transplant Imaging for Liver Complications
Three imaging subtypes are identified in the discovery cohort. These subtypes are not significantly associated with common clinical features, including age, cirrhosis, Child-Pugh score, vascular invasion, distant metastasis, lymph node metastasis, portal vein thrombosis, alpha-fetoprotein level or Barcelona Clinic Liver Cancer stage. However, overall survival differs significantly between the three groups.
The subtypes also show different tumour appearances on imaging. Subtype 2 has the lowest uniformity score, which indicates greater variation within the tumour. Subtype 1 has the highest surface-volume ratio, suggesting more irregular tumour shape. Subtype 2 also has the highest local contrast feature, pointing to a more complex internal tumour structure. Subtype 3 appears more homogeneous by comparison. These imaging differences help separate tumours into groups with distinct clinical patterns.
Imaging Subtypes Predict Immunotherapy Response
The same clustering method is applied to an immunotherapy-treated validation cohort from the First Affiliated Hospital of Sun Yat-sen University. Patients in this cohort have hepatocellular carcinoma confirmed by pathology or imaging, contrast-enhanced computed tomography before treatment and consecutive treatment with anti-PD-L1 or anti-PD-1 monoclonal antibodies according to physician protocol.
The three imaging subtypes show clear differences in overall survival and progression-free survival. Subtype 1 has the most favourable response to immunotherapy, subtype 2 has the poorest response and subtype 3 has an intermediate response. In survival modelling, imaging subtype is the only independent predictor of survival outcomes in the immunotherapy-treated validation cohort. Compared with subtype 1, subtype 2 shows a higher risk of progression and death, while subtype 3 shows a smaller but still significant increase in risk.
The subtype pattern remains broadly consistent between the discovery and validation cohorts. Heatmap profiles, in-group proportion values and pooled clustering support this consistency. Only a small number of patients receive different subtype assignments when direct clustering is compared with pooled clustering. Principal component analysis also shows no significant differences in imaging features between the discovery and validation cohorts across the three subtypes.
Immune Activity Differs Between Subtypes
A gene-based classifier extends the imaging subtype approach to cohorts with gene expression data but without imaging data. The classifier is trained using imaging subtype labels and RNA sequencing data from the TCGA liver hepatocellular carcinoma imaging validation cohort. It is then applied to TCGA and International Cancer Genome Consortium gene validation cohorts. Predicted imaging subtypes show significant differences in overall survival in both gene validation cohorts, supporting a link between imaging patterns and tumour biology.
Gene expression and pathway analyses show immune-related differences between the subtypes. Compared with subtype 2, subtypes 1 and 3 show stronger enrichment of immune-related processes. These include B-cell activity, lymphocyte-mediated immunity, immune response activation and innate immune response. The results indicate that the CT-based groups reflect differences in both adaptive and innate immune activity.
Subtype 2, which has the poorest response to immunotherapy, has the lowest infiltration of activated CD8-positive T cells, activated B-cells and mast cells. It also has the highest infiltration of Th17 helper cells, Th2 helper cells, regulatory T cells, natural killer cells and neutrophils. Cytolytic activity is lower in subtype 2 than in the other groups. Type I and type II interferon responses are also reduced in subtype 2 and increased in subtype 1. Tumour mutational burden does not differ significantly between the subtypes.
Quantitative computed tomography features identify three hepatocellular carcinoma imaging subtypes with different survival patterns, immunotherapy responses and immune signatures. Subtype 1 shows the most favourable response profile, subtype 2 the poorest and subtype 3 an intermediate pattern. The framework links non-invasive imaging findings with immune features such as B-cell activity, lymphocyte-mediated immunity, cytolytic activity and interferon responses. Larger prospective cohorts and further experimental work remain necessary to address scanner variation, feature selection, observer variability and links with histopathological characteristics.
Source: European Radiology Experimental
Image Credit: iStock
References:
Wu S, Peng F, Huang J et al. (2026) Clustering of quantitative CT features identifies HCC subtypes with distinct prognosis and immune signatures. Eur Radiol Exp; 10: 75.
