Scholarlink

<p>Breast cancer (BC) remains a critical global health challenge, necessitating early and precise diagnostic tools. This study uses Raman spectroscopy and sophisticated machine learning to offer a revolutionary automated approach for non-invasive breast cancer detection and staging. The Cumulative Curve Fitting Approximation (CCFA) algorithm is introduced for preprocessing, effectively eliminating noise and baseline drift to enhance spectral clarity. Feature selection is optimized through the Hybrid War Strategy Optimization with Tactical Unit Algorithm (WSO-TUA), which identifies the most informative spectral biomarkers, reducing redundancy and dimensionality. For classification, a Node-Level Sparse Graph Attention Capsule Neural Network (NSGACNN) is developed, which captures both local dependencies and hierarchical relationships in the data, improving feature representation and robustness. Furthermore, the Hyperbolic Sine Optimizer (HSO) is employed for hyperparameter tuning, enhancing convergence and model generalization. Tested on 2,340 blood plasma Raman spectra, the suggested system outperforms conventional models in terms of computing efficiency and diagnostic precision, achieving an accuracy of 99.85%. These results establish a reliable, high-performance framework for biomarker-based breast cancer detection and staging.</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

Raman Spectroscopy and Node-Level Sparse Graph Attention Capsule Neural Network for Biomarker-Based Breast Cancer Detection and Staging from Blood

  • 期刊论文

  • 发表日期：

    2025年8月5日

查看全文
  • K. Baranitharan,
  • L. Arulmozhiselvan,
  • Aanandha Saravanan K,
  • Bharti Sharma

摘要

Breast cancer (BC) remains a critical global health challenge, necessitating early and precise diagnostic tools. This study uses Raman spectroscopy and sophisticated machine learning to offer a revolutionary automated approach for non-invasive breast cancer detection and staging. The Cumulative Curve Fitting Approximation (CCFA) algorithm is introduced for preprocessing, effectively eliminating noise and baseline drift to enhance spectral clarity. Feature selection is optimized through the Hybrid War Strategy Optimization with Tactical Unit Algorithm (WSO-TUA), which identifies the most informative spectral biomarkers, reducing redundancy and dimensionality. For classification, a Node-Level Sparse Graph Attention Capsule Neural Network (NSGACNN) is developed, which captures both local dependencies and hierarchical relationships in the data, improving feature representation and robustness. Furthermore, the Hyperbolic Sine Optimizer (HSO) is employed for hyperparameter tuning, enhancing convergence and model generalization. Tested on 2,340 blood plasma Raman spectra, the suggested system outperforms conventional models in terms of computing efficiency and diagnostic precision, achieving an accuracy of 99.85%. These results establish a reliable, high-performance framework for biomarker-based breast cancer detection and staging.