% 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{Sun:302986,
author = {Y. Sun and H. Dong and W. Jin and H. Xuan and Z. Yuan and
L. Käsmann$^*$ and L. Shen and T. Wang and X. Ye and M.
Zeng},
title = {{E}nhancing the intraoperative identification of high-grade
patterns in invasive lung adenocarcinoma via radiomics.},
journal = {Translational Lung Cancer Research},
volume = {14},
number = {6},
issn = {2218-6751},
address = {[Erscheinungsort nicht ermittelbar]},
publisher = {[Verlag nicht ermittelbar]},
reportid = {DKFZ-2025-01433},
pages = {2145 - 2158},
year = {2025},
abstract = {High-grade patterns (HGPs) are important for surgical
decision-making in patients with invasive lung
adenocarcinoma (IAC), but the sensitivity of intraoperative
frozen section (FS) is not high. Radiomics has the potential
to improve the sensitivity of intraoperative detection. The
purpose of the present study was to evaluate the value of
combining radiomics with FS analysis for predicting HGPs in
patients with clinical T1 (cT1) IAC.Data from a total of 490
patients who were surgically diagnosed with IAC from January
2019 to April 2019 were retrospectively analyzed; the
patients were randomly divided into a training set (n=392)
and a test set (n=98). The presence of HGPs (micropapillary,
solid, and complex glandular patterns) was evaluated
according to the final pathology (FP). Radiomics features
were extracted from thin-slice computed tomography (CT)
images, and feature selection was performed via the mutual
information method and least absolute shrinkage and
selection operator regression algorithm. The radiomics (R),
FS, and radiomics-frozen section (R-FS) models were
established to predict the presence of HGPs in FP. The area
under the receiver operating characteristic (ROC) curve, the
precision-recall curve, the calibration curve, and decision
curve analysis were used to evaluate model performances. The
permutation importance algorithm (PIA) and local
interpretable model-agnostic explanations (LIME) were used
to provide interpretations for the R model. Additionally,
the predictive performance was compared among tumors with
different CT densities.The R and R-FS models outperformed
the FS model, with the R-FS model achieving the best area
under the curve value of 0.907 $(95\%$ confidence interval:
0.830-0.956) in the test set. PIA and LIME determined the
interpretability of outputs from both the overall model and
individual sample perspectives. Among the three models, the
R model performed best in pure ground-glass nodules and
pure-solid tumors.Radiomics could function as a
complementary check to FS to provide a more sensible and
accurate intraoperative identification of HGPs as compared
to the use of FS alone, thus better informing clinical
decision-making.},
keywords = {Radiomics (Other) / frozen section (FS) (Other) /
high-grade pattern (HGP) (Other) / lung adenocarcinoma
(LUAD) (Other) / machine learning (Other) / non-high-grade
pattern (non-HGP) (Other)},
cin = {MU01},
ddc = {610},
cid = {I:(DE-He78)MU01-20160331},
pnm = {899 - ohne Topic (POF4-899)},
pid = {G:(DE-HGF)POF4-899},
typ = {PUB:(DE-HGF)16},
pubmed = {pmid:40673081},
pmc = {pmc:PMC12261232},
doi = {10.21037/tlcr-2025-504},
url = {https://inrepo02.dkfz.de/record/302986},
}