Scholarlink

Abstract <p>Breast cancer is the most commonly diagnosed cancer among women globally, and its early detection is essential for effective treatment and improved patient outcomes. Traditional diagnostic approaches often rely on imaging and statistical analysis techniques, which may not effectively capture complex relationships within clinical data. This study introduces a novel deep learning framework utilizing graph neural networks (GNNs) for breast cancer classification. The proposed model transforms structured data into a graph format, allowing for the representation of inter-feature dependencies through message-passing mechanisms. A weighted K-nearest neighbors algorithm is used to construct the graph, followed by a graph embedding process that encodes both local and global feature relationships. A two-layer graph convolutional network is employed to perform node-level classification, distinguishing between benign and malignant cases. The computational demands of graph construction, embedding, and hyperparameter optimization necessitated the use of high-performance computing (HPC) resources. Experiments were conducted on an HPC cluster featuring multi-core CPUs and GPUs to enable efficient data parallelism and reduced training latency. The parallelized implementation accelerated convergence, supported large-scale parameter tuning, and ensured the model’s scalability for future real-time clinical applications. The proposed model, RBC-GNN(Relation-aware Breast Cancer Classification using GNN), is evaluated using the Wisconsin Diagnostic Breast Cancer (WDBC) dataset. Key performance metrics, including accuracy, precision, recall, F1-score, confusion matrices, AUC–ROC, and loss curves, are used to assess effectiveness. RBC-GNN achieves a classification accuracy of 97.37%, demonstrating its superior performance and potential for clinical application. Comparative analysis with various optimizers confirms the robustness of the proposed approach. The results highlight the synergy between GNN architectures and HPC environments in enabling fast, scalable, and reliable computational pathology. This work underscores the potential of GNN-driven, HPC-accelerated frameworks for intelligent and resource-efficient breast cancer diagnosis in clinical practice.</p> Graphical abstract <p></p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

RBC-GNN: a novel relation-aware graph-based learning framework for breast cancer classification using graph neural network

  • 期刊论文

  • 发表日期：

    2025年11月7日

查看全文
  • Hemali Shah,
  • Smita S Agrawal,
  • Parita Oza,
  • Sudeep Tanwar

摘要

Abstract

Breast cancer is the most commonly diagnosed cancer among women globally, and its early detection is essential for effective treatment and improved patient outcomes. Traditional diagnostic approaches often rely on imaging and statistical analysis techniques, which may not effectively capture complex relationships within clinical data. This study introduces a novel deep learning framework utilizing graph neural networks (GNNs) for breast cancer classification. The proposed model transforms structured data into a graph format, allowing for the representation of inter-feature dependencies through message-passing mechanisms. A weighted K-nearest neighbors algorithm is used to construct the graph, followed by a graph embedding process that encodes both local and global feature relationships. A two-layer graph convolutional network is employed to perform node-level classification, distinguishing between benign and malignant cases. The computational demands of graph construction, embedding, and hyperparameter optimization necessitated the use of high-performance computing (HPC) resources. Experiments were conducted on an HPC cluster featuring multi-core CPUs and GPUs to enable efficient data parallelism and reduced training latency. The parallelized implementation accelerated convergence, supported large-scale parameter tuning, and ensured the model’s scalability for future real-time clinical applications. The proposed model, RBC-GNN(Relation-aware Breast Cancer Classification using GNN), is evaluated using the Wisconsin Diagnostic Breast Cancer (WDBC) dataset. Key performance metrics, including accuracy, precision, recall, F1-score, confusion matrices, AUC–ROC, and loss curves, are used to assess effectiveness. RBC-GNN achieves a classification accuracy of 97.37%, demonstrating its superior performance and potential for clinical application. Comparative analysis with various optimizers confirms the robustness of the proposed approach. The results highlight the synergy between GNN architectures and HPC environments in enabling fast, scalable, and reliable computational pathology. This work underscores the potential of GNN-driven, HPC-accelerated frameworks for intelligent and resource-efficient breast cancer diagnosis in clinical practice.

Graphical abstract