CT-Based Habitat Radiomics for Quantitative Characterization of Intratumoral Heterogeneity and p40 Immunohistochemical Expression
Abstract
Purpose
To develop and validate a CT-based habitat radiomics approach for quantitative characterization of intratumoral heterogeneity and assessment of p40 protein expression derived from immunohistochemistry.
Methods
This retrospective multicenter study included 123 lung squamous cell carcinoma patients, with patients categorized into p40-positive and p40-negative groups based on immunohistochemical assessment of p40 protein expression. Data from two centers were used for model training and validation (n = 95), and data from a third center served as an independent test set (n = 28). Conventional radiomics features were extracted from whole tumor regions. For habitat radiomics, tumors were partitioned into 100 supervoxels using simple linear iterative clustering, followed by Gaussian mixture modeling to identify intratumoral habitats, from which habitat-based radiomics features were derived. Feature selection was performed using t-test, Pearson correlation analysis, and least absolute shrinkage and selection operator (LASSO). Multiple machine learning classifiers were evaluated to construct predictive models. Model performance was assessed using area under the receiver operating characteristic curve (AUC), calibration curves, and decision curve analysis. SHapley additive explanations (SHAP) were used to interpret feature contributions.
Results
A total of 1,000 radiomics features were extracted, including 200 conventional features and 800 habitat-based features. Four features were retained for the conventional radiomics model, while sixteen features were selected for the habitat model. The conventional model achieved AUCs of 0.73, 0.77, and 0.75 in the training, validation, and test sets, respectively. In contrast, the habitat radiomics model demonstrated improved performance, with AUCs of 0.91, 0.81, and 0.83 across the corresponding datasets. SHAP analysis indicated that habitat-based features contributed substantially to biomarker expression prediction.
Conclusion
CT-based habitat radiomics enables accurate noninvasive characterization of intratumoral heterogeneity and p40 immunohistochemical expression, outperforming conventional whole-tumor radiomics. This approach provides a quantitative imaging surrogate for p40-related tumor phenotyping.