Lung cancer remains a leading cause of cancer-related deaths, emphasizing the need for early and accurate detection. While traditional methods like computed tomography (CT) scans and X-rays are widely used, they rely on manual interpretation, leading to variability and errors. To address these limitations, this study evaluates the performance of three advanced deep learning models-Convolutional Neural Networks (CNNs), CSWin Transformers, and RoFormer-in detecting lung cancer from radiological images. The methodology includes a structured pipeline with preprocessing, model training, validation, and testing, using a three-way data split (training: 64%, validation: 16%, testing: 20%). CNNs achieved the highest performance with an accuracy of 98%, recall of 1.00, and F1-score of 0.98, outperforming CSWin Transformers and RoFormer, which achieved accuracies of 94% and 93%, respectively. Preprocessing steps, such as normalization and augmentation, improved model robustness and addressed data challenges. In conclusion, CNNs proved highly effective, particularly in minimizing false negatives, while transformer-based models showed potential for complementary use. Future work should focus on improving model interpretability, data diversity, and hybrid architectures to enhance clinical adoption and patient outcomes.