% IMPORTANT: The following is UTF-8 encoded. This means that in the presence
% of non-ASCII characters, it will not work with BibTeX 0.99 or older.
% Instead, you should use an up-to-date BibTeX implementation like “bibtex8” or
% “biber”.
@ARTICLE{RabascoMeneghetti:309874,
author = {A. Rabasco Meneghetti$^*$ and C. Campani and C. Roux and Z.
I. Carrero and D.-A. Popica and G. Amaddeo and M. Lequoy and
C. Hollande and S. Mouri and M. Wagner and V. Plaforet and
S. Sidali and M. Ronot and M. Rudler and A. Luciani and O.
Sutter and E. Spitzer and H. Regnault and S. El Mouhadi and
V. Ozenne and N. Ganne-Carrié and M. Bouattour and J. C.
Nault and D. Thabut and J. N. Kather$^*$ and M. Allaire},
title = {{D}etection of {E}sophageal {V}arices and {P}rediction of
{H}epatic {D}ecompensation in {U}nresectable
{H}epatocellular {C}arcinoma using {AI}: {AI} {D}etection of
{V}arices and {D}ecompensation.},
journal = {Journal of hepatology},
volume = {nn},
issn = {0168-8278},
address = {Amsterdam [u.a.]},
publisher = {Elsevier Science},
reportid = {DKFZ-2026-00359},
pages = {nn},
year = {2026},
note = {#DKTKZFB9# / #NCTZFB9# / epub},
abstract = {In hepatocellular carcinoma (HCC) with cirrhosis, portal
hypertension (PHT) worsens outcomes.
Esophagogastroduodenoscopy (EGD), current screen for
esophageal varices (EVs), is invasive and delays therapy. We
aimed to develop and externally validate non-invasive models
to detect EV and predict hepatic decompensation (bleeding,
ascites or hepatic encephalopathy), a major cause of HCC
mortality, using routine contrast-enhanced CT (CECT) and
clinical data.This multicenter retrospective study included
489 patients with unresectable HCC treated with
Atezolizumab-Bevacizumab (AtezoBev) from five French
centers, split into a development (n=279) and external
validation (n=210) cohorts. Arterial-phase CECTs were
processed through a Deep Learning pipeline using a
foundation model (HepatoSageCT). Logistic and Cox models
generated clinical models and combined models integrating
the HepatoSageCT scores with key clinical variables for EVs
and hepatic decompensation. Performance was assessed using
AUCROC, sensitivity, specificity, concordance index and
cause-specific hazard ratio.Portosystemic shunts (PSS) at
imaging identified EVs with AUCROC of 0.78, increasing to
0.84 when combined with HepatoSageCT. A decision algorithm
incorporating PSS and HepatoSageCT missed $4.2\%$ of varices
needing treatment, compared to $8.4\%$ when using only PSS
while missing $0\%$ of large EVs. HepatoSageCT predicted
hepatic decompensation in validation (C-Index: 0.73, hazard
ratio: 3.17) with significant stratification (p<0.001)
comparable to a score of ascites, splenomegaly and
HepatoSageCT risk (C-Index: 0.73, hazard ratio: 3.48).
Patients stratified at higher risk of decompensation by
HepatoSageCT also showed significantly (p<0.001) lower
overall survival (OS).HepatoSageCT scores, supplemented with
clinical data, enables accurate non-invasive detection of EV
in AtezoBev-treated unresectable HCC and stratifying
patients based on hepatic decompensation risk, potentially
assisting in reducing unnecessary endoscopies and guiding
prognosis.The present study demonstrates that foundation
models applied to routine CT imaging, when combined with
routinely-collected features such as the presence of
portosystemic shunts, can accurately predict the presence of
esophageal varices and risk of hepatic and first hepatic
decompensation in patients with AtezoBev-treated
unresectable HCC. These findings are particularly relevant
for hepatologists and oncologists, as they highlight a
promising non-invasive tool for timely risk assessment in a
time-sensitive patient population. While prospective
validation is warranted, this approach could support more
personalized management and care of patients with
unresectable HCC.},
keywords = {deep learning (Other) / hepatic decompensation (Other) /
hepatocellular carcinoma (Other) / portal hypertension
(Other) / portosystemic shunt (Other) / varices (Other)},
cin = {DD01 / HD02},
ddc = {610},
cid = {I:(DE-He78)DD01-20160331 / I:(DE-He78)HD02-20160331},
pnm = {899 - ohne Topic (POF4-899)},
pid = {G:(DE-HGF)POF4-899},
typ = {PUB:(DE-HGF)16},
pubmed = {pmid:41679555},
doi = {10.1016/j.jhep.2026.01.021},
url = {https://inrepo02.dkfz.de/record/309874},
}