The tumor immunohistochemical marker ki-67 is an important marker of cell proliferation, and an important prognostic factor for non-small cell lung cancer (NSCLC). CT radiomics features can reflect potential histopathological changes without trauma. The lesions are segmented by a radiomics physician, the 3D reconstructions are performed and 107 radiomics features are extracted. The correlation between CT radiomics features based on NSCLC and ki-67 is analyzed and 45 features are selected. Then we perform factor analysis to further select features. Significantly correlated with ki-67 expression level, nine features are selected. Finally, we use support vector machines (SVM) optimized by particle swarm optimization (PSO) to build a prediction classifier. SVM has a fast convergence speed and strong resistance to overfitting. Meanwhile, PSO has a strong global search capability. The prediction accuracy of ki-67 expression level conducted by PSO-SVM classifier is 91.30%.