Lung cancer is predominantly illustrated as the principal cause of cancer-related deaths globally, especially the diagnosis of late stages creates substantial reductions in survival rate. Recent advancements in artificial intelligence (AI) and medical imaging offer promising avenues for early and accurate detection of pulmonary malignancies. This paper introduces an EfficientNetB0 deep learning architecture used for performing multiclass lung cancer detection through computed tomography scan analysis. The EfficientNetB0 framework was validated, trained and tested on six clinically relevant CT scan image types within a publicly accessible Kaggle database. A combination of transfer learning with complete fine-tuning and customized classification head along with regularization enabled the model to reach a test accuracy of 83.58% macro-average AUC of 0.9492 and a weighted F1-score of 0.85. The testing results demonstrated excellent performance in malignant and normal classes, however have an insufficient ability to identify underrepresented benign cases due to class imbalance effects. This research includes visual diagrams of system architecture together with training performance graphs and a complete metric data examination. The achieved results elucidated EfficientNetB0 as an effective and lightweight backbone solution for computer-aided diagnosis systems used in pulmonary oncology.