Abstract

Background: To evaluate the impact of CT-based deep learning model of hepatic venous pressure gradient (HVPG) on prognosis of hepatocellular carcinoma (HCC) patients treated with transarterial chemoembolization (TACE) and systemic therapy.

Methods: A total of 261 consecutive HCC patients treated with TACE and systemic therapy, and had a contrast-enhanced abdominal CT as part of their pre-surgical work-up, were retrospectively collected between January 2010 and December 2021. A CT-based HVPG Score, whose computed formula was: 17.37-4.91*ln(Liver/Spleen volume ratio) +3.8[If presence of peri-hepatic ascites],was used to diagnose portal hypertension (image-based CSPH, iCSPH for short) with a cut-off value 11.606. The 3D liver and spleen volume were automate calculated by a deep learning segmentation model, and the presence of peri-hepatic ascites was diagnosed by two independent investigators in portal-venous phase CT. Overall survival (OS) as study endpoint was analyzed by Kaplan-Meier and Cox regression.

Results: Among 261 patients, 80(30.7%) were diagnosed with iCSPH by CT-based HVPG Score. The median OS in iCSPH group was significantly shorter than non-iCSPH group (16.9 months vs. 20.7 months, P=0.022). Multivariable analysis indicated that the presence of iCSPH was a negative prognostic factor for OS (adjusted hazard ratio [HR], 1.423, P=0.045).

Conclusion: The segmentation model shows good performance in liver and spleen segmentation in HCC patients, which may help non-invasive HVPG assessment and other CT imaging studies in HCC patients. CT-based HVPG Score was significantly associated with poor outcome and should be taken into consideration when managing HCC patients underwent TACE and systemic therapy.