Decoding intratumoral heterogeneity with spatially resolved PET/CT habitat radiomics for preoperative identification of IASLC grade 3 lung adenocarcinoma: a multicenter study

Poorly differentiated lung adenocarcinoma (LUAD) is generally associated with more aggressive behavior and an unfavorable prognosis. Therefore, we aimed to develop and validate a PET/CT habitat radiomics framework for the preoperative prediction of poorly differentiated LUAD. This retrospective two-center study included 279 patients with pathologically confirmed LUAD. Patients from center 1 were allocated to training ( n = 144) and internal test ( n = 63) sets, and patients from center 2 formed an external test set ( n = 72). All patients underwent preoperative 18 F-FDG PET/CT and surgical resection. CT and PET voxelwise feature maps were partitioned into two and four habitats, respectively. Clinical, conventional radiomics, and habitat features were combined using early- and late-fusion strategies. Model performance was evaluated using discrimination, calibration, and decision-curve analysis. Late1 fusion achieved the highest AUC in the training set (0.835; 95% CI, 0.762–0.899) and internal test set (0.782; 95% CI, 0.661–0.888). In the external test set, Late2 fusion achieved the numerically highest AUC (0.842; 95% CI, 0.738–0.930), followed by Late1 fusion (0.811), conventional radiomics (0.806), the clinical model (0.801), habitat radiomics (0.794), and Early fusion (0.748). No pairwise AUC differences were statistically significant in either the internal or external test set (all P > 0.05). In the external test set, Late2 fusion had an accuracy of 0.750, sensitivity of 0.733, specificity of 0.762, and the lowest Brier score (0.161). Conventional radiomics and multimodal fusion models showed good discrimination for poorly differentiated LUAD in the independent external test set. However, no single fusion strategy consistently outperformed the other models across all cohorts. Prospective multicenter evaluation, calibration assessment, and reproducibility analysis are required before clinical implementation.

Authors

Institutions

Publication Details

Journal
BMC Cancer
Published
2026-11-17
DOI
https://doi.org/10.1186/s12885-026-17379-z
Primary Topic
Radiomics and Machine Learning in Medical Imaging
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Decoding intratumoral heterogeneity with spatially resolved PET/CT habitat radiomics for preoperative identification of IASLC grade 3 lung adenocarcinoma: a multicenter study

Ke Xu, Furui Duan, Wenbin Li, Yanjun Dong et al.
6 links
BMC Cancer
Radiomics and Machine Learning in Medical Imaging
article

Decoding intratumoral heterogeneity with spatially resolved PET/CT habitat radiomics for preoperative identification of IASLC grade 3 lung adenocarcinoma: a multicenter study

Ke Xu, Furui Duan, Wenbin Li, Yanjun Dong, Jianting Man, Xiang Cui, Wen Wang, Yufei Huang, Zhenyong Li, Mingrui Yang, Quanyuan Liu
article en

Abstract

Poorly differentiated lung adenocarcinoma (LUAD) is generally associated with more aggressive behavior and an unfavorable prognosis. Therefore, we aimed to develop and validate a PET/CT habitat radiomics framework for the preoperative prediction of poorly differentiated LUAD. This retrospective two-center study included 279 patients with pathologically confirmed LUAD. Patients from center 1 were allocated to training ( n = 144) and internal test ( n = 63) sets, and patients from center 2 formed an external test set ( n = 72). All patients underwent preoperative 18 F-FDG PET/CT and surgical resection. CT and PET voxelwise feature maps were partitioned into two and four habitats, respectively. Clinical, conventional radiomics, and habitat features were combined using early- and late-fusion strategies. Model performance was evaluated using discrimination, calibration, and decision-curve analysis. Late1 fusion achieved the highest AUC in the training set (0.835; 95% CI, 0.762–0.899) and internal test set (0.782; 95% CI, 0.661–0.888). In the external test set, Late2 fusion achieved the numerically highest AUC (0.842; 95% CI, 0.738–0.930), followed by Late1 fusion (0.811), conventional radiomics (0.806), the clinical model (0.801), habitat radiomics (0.794), and Early fusion (0.748). No pairwise AUC differences were statistically significant in either the internal or external test set (all P > 0.05). In the external test set, Late2 fusion had an accuracy of 0.750, sensitivity of 0.733, specificity of 0.762, and the lowest Brier score (0.161). Conventional radiomics and multimodal fusion models showed good discrimination for poorly differentiated LUAD in the independent external test set. However, no single fusion strategy consistently outperformed the other models across all cohorts. Prospective multicenter evaluation, calibration assessment, and reproducibility analysis are required before clinical implementation.

BMC Cancer
Shandong University of Aeronautics (CN), Weifang People's Hospital (CN), Affiliated Hospital of Qingdao University (CN)
Good health and well-being
Radiomics and Machine Learning in Medical Imaging
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon