Explainable AI-Driven Deep Learning Model Comparison for Lung Cancer Detection and Subtype Classification

Lung cancer continues to be a leading contributor to global cancer mortality, highlighting the need for rapid and reliable diagnostic techniques to enhance patient survival rates. Non-Small Cell Lung Cancer (NSCLC), the most prevalent subtype, requires advanced computational methods to support accurate classification and clinical interpretation. Despite remarkable progress in neural network-based imaging analysis, the opacity of these models often limits their clinical acceptance. This work introduces an explainable neural network framework designed to categorize NSCLC subtypes while ensuring interpretability in its decision process. The model integrates knowledge transfer from pre-trained DenseNet, ResNet50, and VGG16 architectures to distinguish among adenocarcinoma, squamous cell carcinoma, large cell carcinoma, and normal tissue in histopathological scans. To promote transparency, the GradCAM technique is incorporated to generate color-weighted overlays that reveal the image regions most influential in each classification. Experimental findings demonstrate that DenseNet achieved an overall accuracy of 93% with balanced precision, recall, and F1-scores across all subtypes. The interpretive heatmaps confirmed that the model's focus corresponded with pathologically relevant regions, improving both trust and diagnostic utility. The proposed approach highlights the potential of explainable neural networks as dependable tools for computer-assisted lung cancer diagnosis.