Paper Title
Advanced Graph Neural Networks for Lung Cancer Classification: Towards Structurally-Aware Diagnostics

Abstract
Lung cancer is one of the most common tumors worldwide, needing early, non-invasive exact classification for a better prognosis.Traditional approaches, such as radiological imaging and convolutional neural networks (CNNs), excel at spatial pattern identification but fall short in capturing structural relationships in lung tissue. Recent advances in Graph Neural Networks (GNNs) provide a possible alternative by using graph-based representations to understand spatial and morphological interactions in tumor microenvironments. In this study, we investigate GNNs’ performance in categorizing lung cancer subtypes such as adenocarcinoma, large cell carcinoma, and squamous cell carcinoma, as well as normal tissue. Medical pictures were turned into graph structures, with nodes representing local regions and edges indicating spatial adjacency. Our AdvancedGNN architecture improves feature extraction by combining Graph Attention Networks (GAT), Graph Isomorphism Networks (GIN) and GraphSAGE. A comparison with CNN-based models such as ResNet, VGG, and DenseNet shows that GNNs have the potential to exceed in terms of accuracy, computational efficiency, and interpretability. The AdvancedGNN architecture achieved a peak validation accuracy of 85.61%, surpassing DenseNet’s 62.50% with significant improvements in F1 scores and performance-to-parameter ratio. These results validate the structural awareness and robustness of GNNs in diagnosing lung cancer. Keywords: Graph Neural Networks, Lung Cancer Classification, Spatial Adjacency, Spatial Adjacency