In view of an improvement of the performance of the automatic malignant classification model of lung nodules, a benign and malignant lung nodule classification algorithm has thus been proposed. First, 3D CT images of pulmonary nodules are used as a model input; then the CT image features are to be extracted with the dual-path network combined with VGG16, with the residual connection used to capture more high-level and semantic information, and the dense connection used to reduce the complexity of the model as well. The experimental results on the Luna16 dataset show that the ROC of the algorithm can reach as high as 90%, with its algorithm performance much better than the same type of algorithm.