Scholarlink

<p>Visual data processing has become increasingly crucial in the medical field, driven by the growing demand for automated diagnostic systems. The integration of deep learning, a branch of artificial intelligence, with image processing techniques has revolutionized tissue analysis and automated disease classification. Although early implementations faced hardware limitations, recent advancements in high-performance GPUs have facilitated the widespread adoption of deep neural networks in medical imaging. These models automatically extract meaningful features from large datasets, significantly enhancing pattern recognition and diagnostic accuracy. This study focuses on the classification of histopathological breast cancer images from the BreakHis dataset using a multi-magnification approach (40×, 100×, 200×, and 400×). Transfer learning techniques and state-of-the-art convolutional neural network architectures, including ConvNeXt, InceptionNeXt, and EfficientNetV2, were employed to develop an automated system for accurate and early breast cancer diagnosis. A comprehensive classification strategy was designed, performing both binary (benign vs. malignant) and multi-class classification (subtypes of benign and malignant tumors) at each magnification level. Notably, InceptionNeXt and ConvNeXt achieved the highest binary classification accuracy of 99.52% at 100× magnification, whereas ConvNeXt outperformed other models in multi-class classification with 95.24% accuracy at 40× magnification. These results demonstrate the effectiveness of state-of-the-art deep learning models in histopathological image analysis. However, the generalization ability of the models is still influenced by dataset characteristics. This study highlights the potential of deep learning as a reliable decision-support tool in breast cancer diagnosis, emphasizing its complementary role rather than a standalone diagnostic solution.</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

Deep learning-based histopathological classification of breast tumors: a multi-magnification approach with state-of-the-art models

  • 期刊论文

  • 发表日期：

    2025年5月12日

查看全文
  • Gizem Irmak,
  • Ahmet Saygılı

摘要

Visual data processing has become increasingly crucial in the medical field, driven by the growing demand for automated diagnostic systems. The integration of deep learning, a branch of artificial intelligence, with image processing techniques has revolutionized tissue analysis and automated disease classification. Although early implementations faced hardware limitations, recent advancements in high-performance GPUs have facilitated the widespread adoption of deep neural networks in medical imaging. These models automatically extract meaningful features from large datasets, significantly enhancing pattern recognition and diagnostic accuracy. This study focuses on the classification of histopathological breast cancer images from the BreakHis dataset using a multi-magnification approach (40×, 100×, 200×, and 400×). Transfer learning techniques and state-of-the-art convolutional neural network architectures, including ConvNeXt, InceptionNeXt, and EfficientNetV2, were employed to develop an automated system for accurate and early breast cancer diagnosis. A comprehensive classification strategy was designed, performing both binary (benign vs. malignant) and multi-class classification (subtypes of benign and malignant tumors) at each magnification level. Notably, InceptionNeXt and ConvNeXt achieved the highest binary classification accuracy of 99.52% at 100× magnification, whereas ConvNeXt outperformed other models in multi-class classification with 95.24% accuracy at 40× magnification. These results demonstrate the effectiveness of state-of-the-art deep learning models in histopathological image analysis. However, the generalization ability of the models is still influenced by dataset characteristics. This study highlights the potential of deep learning as a reliable decision-support tool in breast cancer diagnosis, emphasizing its complementary role rather than a standalone diagnostic solution.