Related Experiment Video
Updated: Mar 13, 2026
Using Micro-computed Tomography for the Assessment of Tumor Development and Follow-up of Response to Treatment in a Mouse Model of Lung Cancer
Published on: May 20, 2016
Super-resolution and habitat radiomics based computed tomography machine-learning model for prediction of lung
Yanqing Ma1,2, Pingshan Zhao3, Haoran Chen4
1Department of Radiology, The First Affiliated Hospital, College of Medicine, Zhejiang University, Hangzhou, China.
Background:
Early differentiation between invasive adenocarcinoma (IAC) and non-IAC pulmonary nodules is crucial for guiding clinical decision-making. Therefore, this study aimed to distinguish IAC from non-IAC pulmonary nodules using intra-tumor radiomics signatures, habitat radiomics analysis, and a combined nomogram by integrating generative adversarial network (GAN) based super-resolution reconstruction.
Methods:
In this multi-center retrospective study, 858 patients [mean ± standard deviation (SD): 57.635±12.978 years] were enrolled as the training set (Center 1, 501 non-IAC cases vs. 357 IAC cases) and 272 external testing patients (Centers 2 and 3, 183 non-IAC cases vs. 89 IAC cases; mean ± SD: 57.037±11.683 years) were included. Univariate and multivariate analyses were conducted to explore clinical characteristics. Radiomics features were extracted from intra-tumor regions and sub-regions. After feature selection, machine learning models, namely the Intra-Model and Habitat-Model, were developed. A combined nomogram integrating significant clinical factors, intra-tumor radiomics and habitat radiomics was constructed and evaluated using area under receiver operator characteristics curve (AUC), decision curve analysis (DCA), and other quantified metrics.
Results:
The Habitat-Model outperformed Intra-Model (training AUC: 0.893 vs. 0.853; testing AUC: 0.882 vs. 0.875) in predicting IAC invasiveness. The combined nomogram demonstrated an incremental advancement in IAC stratification [training AUC: 0.907 (95% CI: 0.887-0.927); testing AUC: 0.895 (95% CI: 0.849-0.941)], with DCA confirming 28-34% net benefit improvement over single-modality approaches at critical thresholds (10-25% risk). Age (P<0.001) and nodule diameter (P<0.001), along with intra-tumor and habitat radiomics, were identified as key contributing factors.
Conclusions:
The spatially resolved habitat radiomics model exhibited higher discriminative accuracy than the classical intra-tumor radiomics model. The combined nomogram framework, which integrated intra-tumor radiomics, habitat radiomics, and significant clinical biomarkers, achieved state-of-the-art performance in IAC stratification. This framework provides a robust tool for precision therapeutic decision-making in pulmonary nodule management.
