AUTHOR=Dad Imam , He JianFeng , Baloch Zulqarnain TITLE=Graph-based analysis of histopathological images for lung cancer classification using GLCM features and enhanced graph JOURNAL=Frontiers in Oncology VOLUME=Volume 15 - 2025 YEAR=2025 URL=https://www.frontiersin.org/journals/oncology/articles/10.3389/fonc.2025.1546635 DOI=10.3389/fonc.2025.1546635 ISSN=2234-943X ABSTRACT=Lung cancer remains a leading cause of global cancer mortality, demanding precise diagnostic tools for accurate subtype classification. This paper introduces a novel Enhanced GraphSAGE (E-GraphSAGE) framework that integrates graph-based deep learning (GBDL) with traditional image processing to classify lung cancer subtypes—Adenocarcinoma (ACA), Squamous Cell Carcinoma (SCC), and Benign Tissue (BNT)—from H&E-stained Whole-Slide Images (WSIs). Our methodology leverages Gray-Level Co-occurrence Matrix (GLCM) features to quantify tissue texture, constructs a Sparse Cosine Similarity Matrix (SCSM) to model spatial relationships, and employs DeepWalk embeddings to capture topological patterns. The E-GraphSAGE architecture optimizes neighborhood aggregation, incorporates dropout regularization to mitigate overfitting, and utilizes Principal Component Analysis (PCA) for dimensionality reduction, ensuring computational efficiency without sacrificing diagnostic fidelity. The model is validated on multicell Lymphocytic cancer classification of Diffuse Large B-cell lymphoma (DLBCL), Follicular Lymphoma (FL) and Small Lymphocytic Lymphoma (SLL), experimental results demonstrate superior performance, achieving 96% training accuracy and 90% validation accuracy, with an F1-score of 0.91 and AUC-ROC of 0.95 (DLBCL), 0.92 (FL), and 0.89 (SLL). Comparative analysis against state-of-the-art models (GAT, GCN, ResNet-50, ViT) reveals our framework’s dominance, attaining an overall accuracy of 0.90, F1-score of 0.905, and macro-average AUC-ROC of 0.93. While maintaining 25.7 sec/slide inference speed—significantly faster than competing methods. This study advances computational pathology by unifying Graph Neural Networks (GNN) with interpretable feature engineering, offering a scalable, efficient solution for cancer subtype classification. The framework’s ability to model multi-scale histopathological patterns—from cellular interactions to tissue architecture—positions it as a promising tool for clinical decision support, enhancing diagnostic precision and patient outcomes in hemato-pathology.