Lung tissue segmentation is vital in computer-assisted diagnosis, and many studies have been devoted to this field. However, there are still challenges because of redundant information in CT images and the deformation of organ tissue. In this study, we constructed a fully convolution neural network-based model called LTS-Net to segment lung tissue from CT images. First, to improve efficiency and guarantee segmentation accuracy, the network model uses three downward max-pooling layers and three up-sampling layers. The number of convolution channels increases exponentially with each down-sampling step, thereby resulting in fast feature extraction. Furthermore, each convolutional layer follows an ReLU and a batch normalization layer to maintain robustness. Finally, we analyzed the influence of the receptive field size for lung tissue segmentation to provide a trade-off between accuracy and efficiency. To evaluate the performance of LTS-Net, we constructed a CT image dataset at West China Hospital of Sichuan University. The results demonstrated the constructed model's performance: LTS-Net obtained a new state-of-the-art result of 0.992 Dice coefficient on the dataset.