Differentiating breast cancer lung metastasis from primary lung cancer using clinical and computed tomography radiomics features.

BACKGROUND: Breast cancer is the most common malignancy in women globally and a leading cause of cancer-related death. This study aimed to differentiate breast cancer lung metastasis (BCLM) from primary lung cancer (PLC) in patients with breast cancer by integrating clinicopathological characteristics and computed tomography (CT) radiomics to develop and validate a predictive model. METHODS: A total of 158 patients with breast cancer who were diagnosed with lung metastasis or PLC between 2013 and 2023 were retrospectively included. The patients were randomly assigned to a training group (n=111) and a testing group (n=47). Clinical data and CT images were collected. Independent predictors were identified using logistic regression analysis. Radiomics features were extracted from intratumoral, peritumoral, and combined regions using three-dimensional (3D) Slicer software. RESULTS: There were 100 patients diagnosed with PLC, and 58 were diagnosed with BCLM. Patients in the BCLM group demonstrated higher tumor (T) stage, node (N) stage, human epidermal growth factor receptor 2 positivity, Ki-67 levels, and serum carbohydrate antigen 153 concentrations compared with those of the PLC group. The clinical model yielded an area under the curve (AUC) of 0.704 in the training group and 0.760 in the testing group. The radiomics models achieved AUC values ranging from 0.742 to 0.952. The combined model demonstrated AUC values of 0.959 in the training group and 0.946 in the testing group. CONCLUSIONS: A combined model incorporating clinical and radiomics features demonstrated high predictive performance for distinguishing BCLM from PLC, supporting improved diagnostic differentiation and clinical decision making.