Scholarlink

<p>Lung cancer remains one of the most common and deadly cancers worldwide, emphasizing the need for early and accurate diagnostic methods to improve patient outcomes. Histopathological images play a crucial role in diagnosis, but current models often fall short due to limited feature extraction capabilities, lower accuracy, and lack of interpretability. Moreover, identifying visual similarities between benign and malignant cells in these images poses a significant challenge, as subtle differences are often difficult to detect. Traditional deep learning models struggle to capture the complex and nuanced patterns in histopathological images, leading to suboptimal performance and hindering effective clinical decision-making. This study introduces X-ResViNet, an explainable fusion model using ResNet-50 and Vision Transformer (ViT-B/16) with attention mechanisms for accurate lung cancer diagnosis on the LC25000 dataset. X-ResViNet merges ResNet-50 and ViT to gather detailed information, using the Convolutional Block Attention Module (CBAM) to sharpen focus and Grad-CAM to help explain the results. This approach addresses key limitations of current diagnostic models by improving accuracy and interpretability. The implementation includes image pre-processing with grayscale conversion and Hematoxylin and eosin (H&amp;E) staining for cellular differentiation. The features are processed by ResNet-50 and ViT to extract local and global information, CBAM to enhance relevant features, and Grad-CAM to visualize decision-making regions. X-ResViNet achieved 99.21% accuracy on the validation set and 98.66% on the test set, setting a new benchmark for automated lung cancer detection and showing significant advancements over traditional models.</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

X-ResViNet: an explainable ResNet and vision transformer fusion model with attention mechanisms for enhanced diagnosis of lung cancer from histopathological images

  • 期刊论文

  • 发表日期：

    2026年2月7日

查看全文
  • D. Manju,
  • V. Sudarsan,
  • S. V. N. Sreenivasu,
  • Naga Malleswara Rao Purimetla,
  • Jeevan Jala,
  • Desidi Narasimha Reddy

摘要

Lung cancer remains one of the most common and deadly cancers worldwide, emphasizing the need for early and accurate diagnostic methods to improve patient outcomes. Histopathological images play a crucial role in diagnosis, but current models often fall short due to limited feature extraction capabilities, lower accuracy, and lack of interpretability. Moreover, identifying visual similarities between benign and malignant cells in these images poses a significant challenge, as subtle differences are often difficult to detect. Traditional deep learning models struggle to capture the complex and nuanced patterns in histopathological images, leading to suboptimal performance and hindering effective clinical decision-making. This study introduces X-ResViNet, an explainable fusion model using ResNet-50 and Vision Transformer (ViT-B/16) with attention mechanisms for accurate lung cancer diagnosis on the LC25000 dataset. X-ResViNet merges ResNet-50 and ViT to gather detailed information, using the Convolutional Block Attention Module (CBAM) to sharpen focus and Grad-CAM to help explain the results. This approach addresses key limitations of current diagnostic models by improving accuracy and interpretability. The implementation includes image pre-processing with grayscale conversion and Hematoxylin and eosin (H&E) staining for cellular differentiation. The features are processed by ResNet-50 and ViT to extract local and global information, CBAM to enhance relevant features, and Grad-CAM to visualize decision-making regions. X-ResViNet achieved 99.21% accuracy on the validation set and 98.66% on the test set, setting a new benchmark for automated lung cancer detection and showing significant advancements over traditional models.