Lung cancer is a dominant cause of cancer mortality in the world. Early diagnosis which should start from the expression of estrogen receptor may well result in better treatment and patient surviving. The early diagnosis of lung cancer may greatly benefit from artificial intelligence techniques. For lung cancer diagnosis, researchers propose various learning-based methods. Segmentation however has been hard due to the variations in size and shape and surrounding tissue complexity. Therefore, we suggest that the above enhanced combined intelligence system tested in present study may be refined to predict lung cancer. The main aim of this study is to develop advanced automated methods for diagnosis and classification of lung cancer in CT through the aid of artificial intelligence. Lobe segmentation, nodule extraction, and cancer/non-cancer classification are the typical steps involved in the process. It is based on the hybrid spiral optimisation intelligent-generalised rough set (SOI-GRS) to choose the optimal CT imaging features. It is suggested to use a 3-stage based modified U-Net model for lobe separation and nodule identification in lung cancer classification. We start off by segmenting the lobe on the CT slices using an adapted-U net architecture. In the second stage, we apply the same modified U-Net to predict the potential nodules, we use the same mask and label. 3) In Step3, candidate nodules are classified into cancer and noncancer using MobileNet. The classification of lung cancer, the extraction of candidate nodules and the segmentation of the lobe have obtained promising experimental results on the open-source LUAN16 dataset. The proposed model contributed with a recall, F1-score, accuracy and precision of about 98–99% in classifying lung cancers.