Abstract
OBJECTIVES: Platinum-based chemotherapy for advanced lung cancer is frequently challenged by platinum resistance, and effective early identification methods remain lacking. Although previous studies have demonstrated the importance of imaging features and clinical laboratory data in identifying platinum resistance in lung cancer, differences in platinum resistance among distinct CT imaging subtypes and their predictive value have not been fully elucidated. This study aims to establish a CT radiomics-based lung cancer subtype classification model, to explore differences in platinum resistance, baseline characteristics, and clinical laboratory data across radiomic subtypes, and to further evaluate the predictive value of clinical laboratory data, radiomic features, and subtype information for platinum resistance. METHODS: This retrospective cohort study included 684 patients with histopathologically confirmed lung cancer who were treated at Xiangya Hospital of Central South University between January 2011 and June 2025. Patients were aged 21-80 years (56±9 years). All patients received standard platinum-based chemotherapy. Baseline information (gender, age, smoking history, pathological subtype, and platinum resistance status) and clinical laboratory data (comprehensive biochemical panel, coagulation function, tumor markers, and routine blood parameters) were collected. Pretreatment CT images were obtained, and 3-dimensional region of interest (ROI) were manually delineated layer by layer along tumor boundaries. A total of 1 228 high-throughput radiomic features were extracted. The optimal number of clusters was determined using the silhouette coefficient, and radiomic phenotypes were identified via hierarchical clustering. Variables showing significant differences between subtypes were screened, and binary logistic regression was applied to quantify the contribution of each variable to platinum resistance. 4 clinical laboratory models were constructed: univariate analysis, multivariate analysis, least absolute shrinkage and selection operator (LASSO) regression, and an all-variable model. Diagnostic performance was assessed using the area under the receiver operating characteristic (ROC) curve (AUC), and clinical net benefit was evaluated using decision curve analysis (DCA). Additional predictive models incorporating radiomic features, clinical laboratory data, radiomic features plus clinical laboratory data, radiomic features plus subtype category, clinical laboratory data plus subtype category, and a combined model integrating clinical laboratory data, radiomic features, and subtype category were further developed and compared for their ability to predict platinum resistance. RESULTS: Radiomics-based clustering divided lung cancer patients into an "imaging-physiological homeostasis subtype" and an "imaging-physiological disequilibrium subtype." Statistically significant differences were observed between the 2 subtypes in platinum resistance status, selected biochemical indicators (carbon dioxide, serum creatinine, creatine kinase, myoglobin), and tumor markers (carcinoembryonic antigen) (all P<0.05). Within specific threshold probability ranges, all 4 clinical laboratory models based on baseline characteristics and laboratory data demonstrated clinical benefit and outperformed the traditional dichotomous classification of "platinum-sensitive" versus "platinum-resistant." Among these models, the all-variable model showed the best predictive performance. Clinical laboratory models exhibited superior performance to radiomics-only models in predicting platinum resistance, and combined models integrating clinical laboratory data with radiomic features showed marked improvement in predictive accuracy. Incorporation of radiomic subtype information further enhanced the predictive ability of multiple models, including clinical laboratory models, radiomic models, and combined clinical laboratory-radiomics models. CONCLUSIONS: CT radiomics can effectively characterize lung cancer heterogeneity. Integrating clinical laboratory data, radiomic features, and radiomic subtype information provides optimal predictive performance for platinum resistance.