Scholarlink

<p>Among all female cancers, breast cancer is the most common one that occurs worldwide. In this case, histopathology is the gold standard for confirmation. Generally, researchers employ Hematoxylin and Eosin (H&amp;E) staining as it is both inexpensive and commonly used in cancer diagnostics. However, Immunohistochemistry (IHC) permits the identification of specific molecular markers, though it involves more tests and is expensive. Usually, computer-aided diagnostic methods that are deep learning-based can extract features only through one network, which limits their generalization ability towards complex histopathological variations. The present research work raises a multi-class classification ensemble framework to overcome the mentioned drawback. The work is aimed at the classification of the first for the breast tissue histology images stained with HE of Human Epidermal Growth Factor Receptor-2 (HER2) status (0+, 1+, 2+, 3+). The authors leverage three pre-trained convolutional neural networks, namely GoogLeNet, DenseNet-201, and WideResNet-50, to extract deep features, which are later combined and classified by a Light Gradient Boosting Machine (LightGBM). Besides, to enhance further classification, the LightGBM hyperparameters are tuned via Neuro Particle Swarm Optimization and Double Adaptive General Variable Neighborhood Search (DA-GVNS). A comprehensive evaluation on the Breast Cancer Immunohistochemistry (BCI) dataset demonstrates that the proposed framework achieves strong and consistent performance, with an accuracy of 97.8%, an F1-score of 97.7%, and an AUC of 99.2%, outperforming existing state-of-the-art methods. These results indicate that the proposed approach demonstrates the feasibility of automated HER2 classification from H&amp;E-stained images as a proof-of-concept, with the potential to support HER2 assessment workflows. Further multi-center validation is required before clinical deployment.</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

A Metaheuristic-Driven Ensemble of CNN and Light-GBM for Multi-Class HER2 Breast Cancer Diagnosis

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年3月5日

查看全文
  • Swathi Gowroju,
  • K Shahu Chatrapati,
  • J. Iqbal,
  • Suresh Babu Baluguri,
  • Ausif Padder,
  • Kaushik Dehingia

摘要

Among all female cancers, breast cancer is the most common one that occurs worldwide. In this case, histopathology is the gold standard for confirmation. Generally, researchers employ Hematoxylin and Eosin (H&E) staining as it is both inexpensive and commonly used in cancer diagnostics. However, Immunohistochemistry (IHC) permits the identification of specific molecular markers, though it involves more tests and is expensive. Usually, computer-aided diagnostic methods that are deep learning-based can extract features only through one network, which limits their generalization ability towards complex histopathological variations. The present research work raises a multi-class classification ensemble framework to overcome the mentioned drawback. The work is aimed at the classification of the first for the breast tissue histology images stained with HE of Human Epidermal Growth Factor Receptor-2 (HER2) status (0+, 1+, 2+, 3+). The authors leverage three pre-trained convolutional neural networks, namely GoogLeNet, DenseNet-201, and WideResNet-50, to extract deep features, which are later combined and classified by a Light Gradient Boosting Machine (LightGBM). Besides, to enhance further classification, the LightGBM hyperparameters are tuned via Neuro Particle Swarm Optimization and Double Adaptive General Variable Neighborhood Search (DA-GVNS). A comprehensive evaluation on the Breast Cancer Immunohistochemistry (BCI) dataset demonstrates that the proposed framework achieves strong and consistent performance, with an accuracy of 97.8%, an F1-score of 97.7%, and an AUC of 99.2%, outperforming existing state-of-the-art methods. These results indicate that the proposed approach demonstrates the feasibility of automated HER2 classification from H&E-stained images as a proof-of-concept, with the potential to support HER2 assessment workflows. Further multi-center validation is required before clinical deployment.