Development and validation of a CT-radiomics model for the diagnosis of lung cancer associated with cystic airspaces: a multicenter study.

backgroundLung cancers associated with cystic airspaces (LCCA) are often subject to clinical misdiagnosis. The aim of this study was to establish a diagnostic model that integrates clinical characteristics and preoperative CT radiomic features, thereby providing a novel non-invasive tool for the diagnosis of LCCA.

methodsPatients with lung cystic lesions from two hospitals were retrospectively reviewed. One hospital's cohort was used for training and internal validation, the other for external validation. Radiomics texture features were extracted from arterial phase images of preoperative contrast-enhanced CT. Clinical, radiomics, and hybrid models were developed by combining clinical and radiological features to differentiate benign from malignant lesions, with model performance evaluated and compared.

resultsA total of 355 patients (median age 59 years [IQR: 52-66]; 57.7% male) were enrolled, including 274 with LCCA and 81 with benign lesions. Age, ground-glass opacity, and Rad-score were independent predictors of lesion type. The hybrid model outperformed other models, with AUCs of 0.87 (95% CI: 0.80-0.94), 0.85 (95% CI: 0.73-0.98), and 0.86 (95% CI: 0.78-0.94) in the training, internal test, and external test sets, respectively. The model showed excellent calibration and significant clinical utility based on decision curve analysis.

conclusionWe have developed and validated a hybrid model that can offer valuable assistance in preoperatively distinguishing between LCCA and benign cystic lesions, thereby facilitating clinical decision-making.