Lung cancer is considered a serious and lethal threat to human life, since about 18.4% of patients pass away due to this disease. It is vital to detect lung cancer in the early stages in order to establish practically possible treatment options for the disease. Computed Tomography (CT)-based deep learning diagnosis, a new approach, is proposed to detect lung cancer in its early stages by concentrating on lung nodules. To detect lung nodules, a depth-separable convolutional neural network-based encoder-decoder deep learning model is proposed. Multi-input and memory-based models were designed to increase the accuracy. The proposed methods would be highly attractive in clinical environments because of robustness, dependability, and accuracy. Lung CT screening using computerized support by radiologists, examined the possible signs of lung cancer is a cliche among the most widely cited illnesses as observed from CT images. The proposed method is implemented with the Luna 16 dataset and achieved state of the art results.