Scholarlink

Breast cancer is a leading global health issue, particularly affecting women, where early and accurate diagnosis is critical for effective treatment. This study proposes an explainable deep learning-based framework for classifying breast tumors using mammographic images. The system integrates a fine-tuned GoogleNet model for binary classification (benign vs. malignant) and employs Grad-CAM to provide visual explanations of the predictions. A comprehensive preprocessing pipeline, including denoising, normalization, and data augmentation, enhances generalizability across multiple public datasets (CBIS-DDSM, BCDR, INbreast). Comparative evaluation with traditional machine learning and deep learning models—including Decision Tree, SVM, DenseNet121, VGG16, and ResNet50—demonstrates that GoogleNet achieves superior accuracy of 96.2%, with an AUC of 98.1%. UNet-based tumor segmentation achieves a Dice Similarity Coefficient (DSC) of 91.8%, while YOLOv5 enables real-time detection with 94.7% accuracy. This interpretable and deployable system offers a robust solution for AI-assisted breast cancer diagnosis in both cloud and edge environments.

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

An Explainable Deep Learning Framework for Mammogram-Based Tumor Classification Using GoogleNet and Grad-CAM

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Vidya G. Nipunge,
  • Jaya H. Dewan

摘要

Breast cancer is a leading global health issue, particularly affecting women, where early and accurate diagnosis is critical for effective treatment. This study proposes an explainable deep learning-based framework for classifying breast tumors using mammographic images. The system integrates a fine-tuned GoogleNet model for binary classification (benign vs. malignant) and employs Grad-CAM to provide visual explanations of the predictions. A comprehensive preprocessing pipeline, including denoising, normalization, and data augmentation, enhances generalizability across multiple public datasets (CBIS-DDSM, BCDR, INbreast). Comparative evaluation with traditional machine learning and deep learning models—including Decision Tree, SVM, DenseNet121, VGG16, and ResNet50—demonstrates that GoogleNet achieves superior accuracy of 96.2%, with an AUC of 98.1%. UNet-based tumor segmentation achieves a Dice Similarity Coefficient (DSC) of 91.8%, while YOLOv5 enables real-time detection with 94.7% accuracy. This interpretable and deployable system offers a robust solution for AI-assisted breast cancer diagnosis in both cloud and edge environments.