Predictive model for differentiating malignant and benign small pulmonary nodules

Background Current evaluation strategies for pulmonary nodules (PNs) emphasize size-based risk stratification. Predictive models are increasingly utilized in the preoperative assessment of PNs. While several models have been proposed to predict malignancy in PNs, few are specifically tailored to small pulmonary nodules (SPNs, ≤ 20 mm). This study entailed the development and validation of a predictive model specifically for differentiating malignant from benign SPNs. Methods A retrospective study was conducted using patient data from two medical centers between January 2019 and December 2024. Patients were separated into training and testing cohorts based on the hospital of origin. All patients underwent computed tomography (CT) and positron emission tomography (PET)/CT examinations. The predictive model was constructed using the training cohort and subsequently validated with the testing cohort. Model performance was also compared against that of an existing SPN predictive model. Results The training set comprised 155 patients (111 malignant and 44 benign), while the testing set included 82 patients (64 malignant and 18 benign). Logistic regression analysis identified lobulation (P = 0.015) and elevated standardized maximum uptake value (SUV _max , P = 0.015) as independent predictors of malignancy. The final prediction model was defined as: X = − 4.446 + 1.305 × lobulation (present = 1, absent = 0) + 0.327 × SUV _max . In the training cohort, the area under the curve (AUC) for the new model was 0.821, compared to 0.715 for the existing model. In the testing cohort, AUCs were 0.853 and 0.701, respectively. Calibration curves demonstrated strong concordance between predicted and actual malignancy probabilities. Conclusions The proposed predictive model demonstrated high diagnostic accuracy differentiating malignant from benign SPNs and may help to reduce the unnecessary biopsies.