Frontiers in Oncology, Год журнала: 2025, Номер 15

Опубликована: Май 30, 2025

Lung cancer remains a leading cause of global mortality, demanding precise diagnostic tools for accurate subtype classification. This paper introduces novel Enhanced GraphSAGE (E-GraphSAGE) framework that integrates graph-based deep learning (GBDL) with traditional image processing to classify lung 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 quantify tissue texture, constructs Sparse Cosine Similarity (SCSM) model spatial relationships, employs DeepWalk embeddings capture topological patterns. The E-GraphSAGE architecture optimizes neighborhood aggregation, incorporates dropout regularization mitigate overfitting, utilizes Principal Component Analysis (PCA) dimensionality reduction, ensuring computational efficiency without sacrificing fidelity. is validated on multicell Lymphocytic classification Diffuse Large B-cell lymphoma (DLBCL), Follicular Lymphoma (FL) Small (SLL), experimental results demonstrate superior performance, achieving 96% training accuracy 90% validation accuracy, an F1-score 0.91 AUC-ROC 0.95 0.92 (FL), 0.89 (SLL). Comparative analysis against state-of-the-art models (GAT, GCN, ResNet-50, ViT) reveals our framework’s dominance, attaining overall 0.90, 0.905, macro-average 0.93. While maintaining 25.7 sec/slide inference speed—significantly faster than competing methods. study advances pathology by unifying Graph Neural Networks (GNN) interpretable feature engineering, offering scalable, efficient solution ability multi-scale histopathological patterns—from cellular interactions architecture—positions it as promising tool clinical decision support, enhancing precision patient outcomes in hemato-pathology.

Язык: Английский