COMBINING AI-processed CECT with clinical data improves the detection of oesophageal varices (OV) and decompensation in unresectable hepatocellular carcinoma (HCC), a 2026 multicentre retrospective study has found.
The study developed HepatoSageCT, a foundation deep leaning model that analyses arterial CECT scans of patients with Atezolizumab-Bevacizumab-treated HCC.
Current Screening Landscape
HCC and portal hypertension (PHT) are closely interconnected complications of cirrhosis. Clinically significant PHT is associated with an increased risk of hepatic decompensation: a major cause of HCC mortality, characterised by bleeding, ascites or hepatic encephalopathy.
The presence of PHT in patients with HCC is associated with a reduced overall survival rate and higher risk of complications.
In populations without HCC, clinically significant PHT can be identified through OVs, usually detected using esophagogastroduodenoscopy. However, this method is invasive and delays therapy.
AI-Processed CECT Outcomes
The study included nearly 500 patients with unresectable HCC from five French centres.
Arterial CECTs were processed using HepatoSageCT. Performance was assessed using AUCROC, sensitivity, specificity, concordance index, and cause-specific hazard ratio.
Portosystemic shunts (PSS), caused by PHT, identified OVs by acting as diagnostic markers with an AUROC of 0.78 at imaging. This increased to 0.84 with HepatoSageCT.
A decisions algorithm missed 8.4% of OVs needing treatment when using only PSS. Incorporating HepatoSageCT with the PSS missed half the percentage of OVs: 4.2%.
HepatoSageCT predicted hepatic decompensation with significant stratification, comparable to a combined approach using both HepatoSageCT and clinical data on ascites and splenomegaly. Patients who were identified as higher risk for decompensation by HepatoSageCT showed significantly lower overall survival.
Limitations
The study was retrospective and limited to French hospitals, creating potential issues with selection bias and generalisability. PSS were also considered equally and introduced as a binary variable regardless of site or quantity. The number of patients with ascites at the outset was limited, restricting analysis to a risk factor description for further decompensation.
Researchers suggested prospective validation and multiphase studies in larger, more diverse populations.
Prognostic Value in Hepatology
The study demonstrated the significant potential of non-invasive models in detecting OVs and predicting hepatic decompensation in patients with Atezolizumab-Bevacizumab-treated unresectable HCC.
A combined approach using clinical data and AI-processed CECT scans offered a framework for improving patient selection and risk stratification in advanced HCC, reducing screening endoscopies and supporting personalised HCC management.
Findings supported the integration of deep learning models for prognostic purposes in hepatology.
Reference
Meneghetti A R et al. Detection of Esophageal Varices and Prediction of Hepatic Decompensation in Unresectable Hepatocellular Carcinoma using AI. J Hepatol. 2026;DOI:10.1016/j.jhep.2026.01.021.
