<p>Lung cancer arises and progresses due to genetic alterations in deoxyribonucleic acid, resulting in aberrant cell proliferation and tumor formation. Early-stage lung cancer often presents without discernible symptoms, necessitating diagnosis through medical imaging. Computed tomography (CT) serves as a primary modality for detecting lung cancer. However, the complex structure of the human lung poses significant challenges for radiologists and clinicians in identifying malignant regions, thus driving the exploration of intelligent diagnostic systems to aid in lung cancer detection. In this approach, feature maps of lung cancer images are integrated across multiple levels of the GoogleNet, creating a semantic hierarchy of lung cancer images. Subsequently, handcrafted texture and brightness features are hierarchically extracted from the semantic image hierarchy and fused together. Following the selection of salient features from lung cancer images through their Shapley values, a k-NN classifier is employed to classify test samples of lung cancer images. The proposed methodology is evaluated for classifying lung cancer into three categories—normal, benign, and malignant—using the “IQ-OTH/NCCD-Lung” and “Kaggle CT scan images” benchmark datasets. The segregation of class categories in the evaluation dataset is visualized using the t-distributed stochastic neighbor embedding statistical technique. Evaluation findings confirm the effectiveness of the proposed approach in classifying lung cancer CT images compared to state-of-the-art methods.</p>