Abstarct: Cancer is one of the deadliest diseases in the world. One of the reasons for the high mortality of this disease has been the wrong diagnosis of the doctor or the delay in diagnosing this disease. Among all cancers, lung cancer is difficult to diagnose due to its uneven and uncoordinated structure. With the advancement of science and technology, the diagnosis of this type of disease can be taken out of the traditional mode and done with automatic and semi-automatic systems. In order to have a system that can diagnose the disease without human intervention, deep learning techniques must be used. This implementation consists of two very important parts: segmentation and classification. The output of the segmentation part is the input of the feature extraction and classification part. There are many problems with segmentation, the most important of which is the uneven structure of the lungs. To eliminate this problem, many methods have been proposed, and the problem with most of these methods is the high level of false positives. In this research, a new U-Net network structure has been used for segmentation, which has had the best segmentation result to date. After this step, to extract the feature, the convolutional neural network is used, which is a suitable structure for extracting the visual features. Then, gradient amplification is used to increase the classification accuracy and the output of the convulsive neural network is connected to another network. In summary, to diagnose lung cancer, pre-processing images were performed and then lung images were segmented using the BCDU-Net neural network. Then, using three-dimensional convolutional neural network and with changes on the laxyers of this network, feature extraction is performed and in the last step, the fully connected laxyer is connected to the XGBOOST network to finally classify with high accuracy.