Abstract:
Lung cancer is one of the deadliest diseases worldwide, making its early and precise detection vital for improving patient survival outcomes. This research presents a comparative evaluation of twenty-five CNN-based deep learning architectures for lung cancer classification using CT scan images. To ensure a fair and detailed analysis, multiple performance metrics such as Precision, Recall, F1-score, MSE, MAE, RMSE, Confusion Matrix, and AUC-ROC were employed. Each model was assessed not only on its classification accuracy but also on its ability to correctly identify class labels and minimize misclassifications. Among 25 models, SE-DenseNet201 emerged as the most effective, achieving the highest accuracy of 99.72% and outperforming all other architectures. The results highlight the strong potential of CNN-based models, with SE-DenseNet201 in particular showing excellent promise for supporting accurate lung cancer prediction and classification.
