Abstract: Aiming at the characteristics of lung CT data with spatial information, we proposed a twostage method based on deep learning for the detection and classification of pulmonary nodules in CT images. Two 3D convolution networks were used to effectively learn nodule features. For the detector part of the method, a 3D semantic segmentation model was designed for nodule detection with a UNetlike encoderdecoder structure to predict the location, size and semantic mask of the nodules. For the classifier part, a 3D dual path network was used to summarize and contract features, and the classification results were given. In order to fully utilize the feature in the original data, the results of the detector were used to sample and mask the original data, and a spatial pyramid pooling layer was added before the classifier to obtain a consistent input scale. The experimental results on public datasets show that the proposed deep learning approach has good performance in the detection and classification of pulmonary nodules in CT images.

Key words: pulmonary nodules, detection, classification, deep learning