A Deep Learning Strategy for Accurate Lung Cancer Subtype Classification Using Convolutional Neural Networks

Cancer-related deaths worldwide are dominated by lung cancer, which requires early diagnosis and treatment for effective treatment. Lung cancer diagnosis accuracy has improved with the use of CT scans combined with deep learning techniques. In this study, a custom EfficientNetB7 architecture was proposed, built upon a pre-trained Convolutional Neural Network (CNN), to categorize chest Computed Tomography (CT) scan images into four categories: adenocarcinomas, large cell carcinomas, squamous cell carcinomas, and normal lung tissue. To evaluate the proposed EfficientNetB7 architecture, metrics such as accuracy, precision, recall, F1-score, and AUC were used, comparing its performance with DenseNet121 and ResNet50, two other advanced deep learning models. Based on the results, the custom EfficientNetB7 outperformed DenseNet121 and ResNet50, achieving a higher accuracy rate of $96.48 \%$, precision rate of $\mathbf{9 6. 2 \%}$, recall rate of $\mathbf{9 6. 2 \%}$, F1 score of $\mathbf{9 6. 2 5 \%}$, and AUC score of $99.5 \%$. This demonstrates its effectiveness as a model for accurately classifying lung cancer, even in cases with smaller nodules. The results of this study show that deep learning models can help develop improved tools for early lung cancer identification.