Scholarlink

One of the most serious illnesses that affect women and can even be fatal is Breast Cancer (BC). It has a major impact on their emotional and physical well-being. Breast calcifications or abnormal tissue growth are major health concerns contributing to BC. Early diagnosis and treatment can significantly lower the death rate and are crucial for survival rates. Medical imaging techniques, particularly ultrasound, are widely used due to their non-invasive nature and effectiveness in identifying abnormal tissues. In this research paper, we develop a Breast Cancer detection approach using pre-trained models such as ResNet, GoogleNet, DenseNet, VGG16 and Xception leveraging Transfer Learning (TL). Deep Learning models with the TL concept have proven quite effective in Breast Cancer detection in recent years. TL is used to utilize the knowledge gained from large-scale image recognition tasks to increase the accuracy of cancer detection and classification in the limited ultrasound imaging domain. This paper compares five deep learning models based on the BUSI dataset of breast cancer, which is available publicly. The dataset consists of 780 images in PNG format, from which 210 images are cancerous, 437 are non-cancerous, and 133 are normal. However, GoogleNet’s transfer learning performance is better than that of other existing models. The accuracy for the test and validation were 99.15% and 99.37%, respectively.

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

Performance Evaluation of Deep Learning Models Using Transfer Learning on Ultrasound Images for Breast Cancer Detection

  • 章节

  • 发表日期：

    2026年

查看全文
  • Laxmi Yadav,
  • Girish Chandra,
  • Divakar Yadav

摘要

One of the most serious illnesses that affect women and can even be fatal is Breast Cancer (BC). It has a major impact on their emotional and physical well-being. Breast calcifications or abnormal tissue growth are major health concerns contributing to BC. Early diagnosis and treatment can significantly lower the death rate and are crucial for survival rates. Medical imaging techniques, particularly ultrasound, are widely used due to their non-invasive nature and effectiveness in identifying abnormal tissues. In this research paper, we develop a Breast Cancer detection approach using pre-trained models such as ResNet, GoogleNet, DenseNet, VGG16 and Xception leveraging Transfer Learning (TL). Deep Learning models with the TL concept have proven quite effective in Breast Cancer detection in recent years. TL is used to utilize the knowledge gained from large-scale image recognition tasks to increase the accuracy of cancer detection and classification in the limited ultrasound imaging domain. This paper compares five deep learning models based on the BUSI dataset of breast cancer, which is available publicly. The dataset consists of 780 images in PNG format, from which 210 images are cancerous, 437 are non-cancerous, and 133 are normal. However, GoogleNet’s transfer learning performance is better than that of other existing models. The accuracy for the test and validation were 99.15% and 99.37%, respectively.