CT-based radiomics for predicting brain metastases as the first failure in patients with curatively resected locally advanced non-small cell lung cancer

PurposeBrain metastasis (BM) is the primary first failure pattern in patients with curatively resected locally advanced non-small cell lung cancer (LA-NSCLC). It not yet possible to accurately predict occurrence of BM. The purpose research develop and validate a prediction model BM-free survival based on radiomics characterising lesions combined clinical characteristics LA-NSCLC.MethodsThis study consisted 124 stage IIB-IIIB NSCLC our institution between January 2014 June 2018. Patients were randomly divided into training validation cohorts using 4:1 ratio. Radiomics features selected from chest CT images before surgery. A signature was constructed LASSO algorithm cohort. Clinical developed Cox proportional hazards model. clinical, radiomics, integrated nomograms constructed. performance models assessed its discrimination, calibration, utility.ResultsThe significantly associated overall discrimination nomogram, C-indexes 0.889 (0.872–0.906, 95 % CI) 0.853 (0.788–0.918, cohorts, respectively, better than nomogram (p < 0.0001 for cohort, p = 0.0008 cohort). Compared also improved varying degrees, but apparent cohort 0.0007 0.0554 calibration curve decision analysis demonstrated that exceeded or predicting survival.ConclusionsCompared nomograms, predictive improved. most suitable LA-NSCLC.