Todayās healthcare system relies heavily on medical imaging to carry out non-invasive diagnostic treatments. For clinical analysis, it entails the development of graphical and functional models of the human body and organs. In order to diagnose non-small cell lung cancer, the Multi-resolution patch (MSP) approach is introduced in this study. The model can enhance detection performance by utilizing an atrous convolution network, however caution must be used while selecting the atrous rate. More data from many institutions is needed to improve the generalization of the model because the validation CT data was only collected at one center; this dataset included lung CT imaging data from healthy individuals. The outcome indicates that the modelās performance can be further enhanced by include data from healthy individuals in the training process. Additionally, two sets of experiments show the value of the pre-processing module and the superiority of the suggested network.