Automatic identification of lung disease is a difficult task for researchers due to image noise that can compromise the image quality of the cancer and reduce its performance. Thresholding, lung image quantization, lung image improvement, and noise removal are all important elements of lung cancer pre-processing to improve the quality of the input image. Image denoising is an important preprocessing activity to reduce noise while preserving edges and other detailed features as much as possible before further image preparation such as feature extraction, segmentation, and image analysis images. By reducing misclassification, this work aims to improve the quality of lung imaging and lung cancer diagnosis. The usage of two-level pre-processing techniques and the CNNDFO strategy in this article was used to evaluate lung CT images to predict lung cancer. Lung CT scans were first obtained from the LIDC-IDRI dataset. This dataset contains 1018 lung images divided into 718 training images and 300 test images After that, a Gaussian filter was used to improve the image quality by replacing the pixel with a gaussian distribution method. After improving the rendering of the image, the watershed method was used to isolate the damaged area. A cluster was established to extract spectral-related features based on Euclidean measures. Using the CNNDFO algorithms were used to train and classify the characteristics, and they were able to accurately predict malignance up to 98% of recall and precision accuracy.

Pre-processing, Gaussian Filter, Watershed Segmentation, Convolutional Neural Networks (CNN), Dragonfly Optimization (DFO).