Preoperative Risk Stratification of Lymph Node Metastasis in cT1 Lung Adenocarcinoma Using CT Radiomics and Habitat Analysis

Abstract Background Accurate preoperative assessment of lymph node metastasis (LNM) is important for risk stratification and surgical planning in patients with cT1 lung adenocarcinoma. This study aimed to develop and externally validate a multi-regional computed tomography (CT) radiomics model incorporating intratumoral, peritumoral, and habitat information for preoperative prediction of LNM. Methods A total of 172 patients with cT1 invasive lung adenocarcinoma from one institution were retrospectively included and randomly divided into a training cohort (n = 120) and an internal validation cohort (n = 52). An independent cohort of 63 patients from a second institution was used for external validation. Radiomic features were extracted from the intratumoral region, a 3-mm peritumoral region, and two habitat subregions generated using K-means clustering. After feature selection, a multi-regional radiomic signature (Radscore) was constructed and combined with preoperative clinical variables using logistic regression. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), sensitivity, specificity, accuracy, positive predictive value (PPV), and negative predictive value (NPV). Results Nineteen radiomic features were retained in the final radiomic signature. In the internal validation cohort, the clinical, radiomic, and combined models achieved AUCs of 0.758, 0.844, and 0.852, respectively. Using a prespecified training-derived threshold of 0.4036, the combined model achieved a sensitivity of 65.0%, specificity of 87.5%, and NPV of 80.0% in the internal validation cohort. In the independent external cohort, the combined model achieved an AUC of 0.837 (95% CI: 0.731–0.930), with an accuracy of 77.8%, sensitivity of 77.3%, specificity of 78.0%, PPV of 65.4%, and NPV of 86.5%. The Radscore was significantly associated with IASLC pathological grade and visceral pleural invasion (both P < 0.001). Conclusions The multi-regional CT radiomics model incorporating intratumoral, peritumoral, and habitat information demonstrated good discrimination for preoperative LNM risk stratification in cT1 lung adenocarcinoma and maintained stable performance in an independent external cohort. The model may provide complementary quantitative information for preoperative risk assessment, although further prospective validation is warranted before clinical implementation.