Lung cancer is one of the most prevalent cancers in the world. It is a particular illness that gets out of hand. and creates aberrant lung cell growth. Deoxyribonucleic acid (DNA) mutation caused by numerous genetic reasons causes these cells to behave differently from other normal cells. However, cancer-related mortality can be decreased with early diagnosis and treatment of patients. In return the use of convolutional neural networks (CNN) in the field of medical imaging diagnosis is widespread, however these networks have drawbacks, including slow training speed and poor diagnostic accuracy due to the size of the input matrix to the algorithm. This article proposes a CNN optimization method based on reducing the computational cost of the CNN. The process optimizes the initial CNN parameters due to a strong segmentation step that came before the classification step, which results in condensed regions at the start of CNN. The model was developed and applied to the analysis and diagnosis of medical imaging data for lung cancer on the basis of this optimization approach. Experimental results on 1190 images show that the suggested method performs more diagnostically than CNN when applied directly to pre-segmented images. The suggested method achieves a diagnostic accuracy of 97.06%, and it only takes 0.2481 seconds to diagnose 1190 digital tomography of the human head image datasets