Multi-Class Classification of Lung Cancer Using Streamlined Convolutional Neural Networks

Lung cancer remains one of the leading causes of cancer-related morbidity and mortality worldwide, highlighting the importance of accurate and timely diagnosis. Medical imaging, particularly Computed Tomography (CT), plays a pivotal role in lung cancer detection due to its ability to capture high-resolution, cross-sectional views of lung tissue. However, interpreting CT scans manually is prone to inter-operator variability and fatigue. This paper proposes a deep learning approach, Convolutional Neural Network (CNN), to classify lung CT images into four categories: normal, adenocarcinoma, squamous cell carcinoma, and large cell carcinoma. Our proposed method relies on streamlined hyperparameters, including batch normalization, dropout, L2 regularization, and data augmentation strategies, to handle imbalanced classes and improve generalization. We conducted experiments using varying image resolutions and train-validation-test splits for model performance evaluation. The best model achieved a 0.96 classification accuracy rate at 192x192 image size with macro average precision, recall, specificity, and F1-score of 0.96, 0.96, 0.99 and 0.96, respectively. Our research findings underscore the potential of streamlined CNN architectures to achieve robust model performance comparable to more computationally demanding methods, thus facilitating broader adoption in medical imaging applications. This study also highlights the impact of different image resolutions and parameter choices, offering a practical solution for clinicians and data scientists with limited resources.