<p>The morphology of benign and malignant solitary pulmonary lesions sometimes overlaps, making the differentiation difficult. This research aimed to create a radiomics-based prediction model using dual-phase ¹⁸F-fluorodeoxyglucose positron emission tomography-computed tomography (¹⁸F-FDG PET/CT) for noninvasive classification of these lesions. A total of 132 patients with solitary pulmonary lesions were included. CT, routine PET (PET₁), delayed PET (PET₂) and clinical data were acquired. Five combinations of radiomic features (CT, CT + PET₁, CT + PET₂, CT + PET₁ + PET₂, CT+(PET₂-PET₁)/PET₁) were analyzed. Feature selection used eight methods, and the top ten ranked features were retained based on their weight coefficients. Seven classifiers were used to construct models. The receiver operating characteristic (ROC) curves of the five optimal radiomics models for solitary pulmonary lesions were compared. The optimal CT+(PET₂-PET₁)/PET₁ model achieved the highest AUC of 0.898 (95% CI: 0.828–0.968), compared to the optimal CT (0.828, 95% confidence interval [CI]: 0.754–0.902), CT + PET₁ (0.858, 95% CI: 0.785–0.931), CT + PET₂ (0.867, 95% CI: 0.796–0.938), and CT + PET₁ + PET₂ (0.868, 95% CI: 0.798–0.939) models. Based on dual-phase ¹⁸F-FDG PET/CT for radiomic analysis, the optimal CT+(PET₂-PET₁)/PET₁ model demonstrated promising diagnostic efficacy and can be a clinical diagnostic tool to distinguish between benign and malignant solitary pulmonary lesions.</p>