Development and validation of a prediction model for early-stage lung cancer based on circulating tumor DNA methylation and radiomics

Although circulating tumor DNA (ctDNA) methylation and radiomics have demonstrated strong potential in early lung cancer diagnosis, their complementary value remains largely unexplored. This study aims to develop a nomogram model combining ctDNA methylation and radiomics features, offering a novel strategy for the early identification of lung cancer. A total of 174 patients with lung nodules who underwent surgical resection were enrolled. Clinical data were collected, radiomic features were extracted, and peripheral blood ctDNA methylation levels ( SHOX2 , RASSF1A , and PTGER4 loci) were detected. Elastic net regression combined with 5-fold cross-validation was employed to select key radiomic features and construct the radiomics score (Rad-score). Multivariable logistic regression and the Akaike Information Criterion (AIC) were used for variable selection to construct three predictive models: clinical-methylation, clinical-radiomics, and a combined model (clinical-methylation-radiomics). The performance of the models was evaluated using the area under the curve (AUC) and decision curve analysis (DCA), and compared with the classical Mayo model. Plasma ctDNA methylation levels of SHOX2 , RASSF1A , and PTGER4 were significantly higher in patients with lung cancer than in those with benign lung nodules ( P < 0.01). Five key radiomic features were identified using elastic net regression to construct the Rad-score. The combined clinical-methylation-radiomics model exhibited excellent diagnostic performance, achieving AUCs of 0.937 in the training set and 0.847 in the validation set, significantly outperforming the Mayo model. The optimal cutoff value for the nomogram based on this model was 137.165 (sensitivity 89.5%, specificity 87.0%), where a score ≥ 137.165 indicated a high risk of malignancy. The prediction model integrating clinical characteristics, ctDNA methylation, and radiomics demonstrated excellent discrimination and clinical utility, which can effectively assist in the early detection of lung cancer. Not applicable.