Habitat Model Based on Lung CT for Predicting Brain Metastasis in Patients with Non-Small Cell Lung Cancer

Background: In lung cancer, the occurrence of brain metastasis (BM) is closely associated with the heterogeneity of the primary lung tumor. This study aimed to develop a habitat-based radiomics model using enhanced computed tomography (CT) lung imaging to predict the risk of BM in patients with non-small cell lung cancer (NSCLC). Methods: A retrospective cohort of 475 patients with NSCLC who underwent enhanced CT of the lungs prior to anti-tumor treatment was analyzed. Volumetric CT images were segmented into tumor subregions via k-means clustering based on voxel intensity and entropy values. Radiomics features were extracted from these subregions, and predictive features were selected using minimum redundancy maximum relevance and least absolute shrinkage and selection operator regression. Two logistic regression models were constructed: a whole-tumor radiomics model and a habitat-based model integrating subregional heterogeneity. Model performance was evaluated via receiver operating characteristic analysis and compared via DeLong’s test. Results: A total of 195 eligible patients with NSCLC were included. The volume of interest of the whole tumor was clustered into three subregions based on voxel intensity and entropy values. In the training cohort (n = 138), the areas under the curve of the clinical model, the whole-tumor model and the habitat-based model were 0.639 (95% confidence interval CI: 0.543–0.731), the whole-tumor model and the habitat-based model were 0.728 (95% confidence interval CI: 0.645–0.812) and 0.819 (95% CI: 0.744–0.894), respectively. The habitat-based model demonstrated superior predictive performance compared with the whole-tumor model (p = 0.022). Conclusions: The habitat-based radiomics model outperformed the whole-tumor model in terms of predicting BM, highlighting the importance of subregional tumor heterogeneity analysis.