Lung cancer is recognized as one of the most significant reasons of mortality in both women and men, causing nearly five million deaths annually. Only in the United States, nearly 200,00 new cases are found every year. The detection and classification process faces several challenges due to the large quantity of data present in CT scan images and blurred boundaries. Similarly, results of manual interpretation and analysis from MR images become more error-prone and time-consuming. In this case, the identification of the small nodules and their locations within the complex structure of the lung makes the early diagnosis more complicated. Here emerges the role of deep learning, which offers efficient and more reliable methodologies. In this paper, we propose an automated method for lung cancer detection with improved efficiency and performance, reducing the overall diagnosis period. Results show that VGG16 and MobileNet outperform the existing architectures in terms of reliability, both achieving 99.0% accuracy. VGG16 also achieved a perfect recall score of 100%, which ensures that there are no false negatives. This indicates that lightweight and robust models can significantly reduce the detection period and assist in early diagnosis of lung cancer.