Scholarlink

<p>In the last decade, thermography has emerged as a harmless and low-cost imaging modality with the potential for breast cancer diagnosis. Alongside this, the use of deep learning for analyzing thermograms has expanded significantly. However, the limited number of labeled thermograms in existing datasets hinders the effective application of deep neural networks. This paper proposes a novel solution to address the challenge of insufficient and imbalanced data by generating synthetic thermograms through a cyclic mapping approach. Such a strategy incorporates the most distinction with the characteristics of the alternative class, during the procedure of constructing fake thermograms belonging to each of the healthy or cancerous classes. Therefore, the deep model which is formed with these thermograms has a better separation potential between these two oppose types of breast tissues. Applying the proposed method on the thermograms of the DMR-IR dataset indicates a significant improvement in the separation of healthy and cancerous cases compared to methods which utilize standard Generative Adversarial Network (i.e., GAN) for production of fake thermograms. The obtained results demonstrate that the proposed scheme may promote characteristics of specificity and accuracy at least 15 and 10 percent compared to existing methods. In terms of sensitivity, a slight improvement up to 1% have also been observed. Furthermore, the proposed scheme reduces the gap between training and validation performance by about one-third and extends the number of effective training epochs by up to four times, helping to prevent premature overfitting in deep neural network training.</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 novel image translation scheme to promote detection of breast cancer in thermograms based on cyclic generative adversarial network

  • 期刊论文

  • 发表日期：

    2026年2月18日

查看全文
  • Seyed Vahab Shojaedini,
  • Mohamad Firouzmand,
  • Mehdi Abedini,
  • Mahsa Monajemi

摘要

In the last decade, thermography has emerged as a harmless and low-cost imaging modality with the potential for breast cancer diagnosis. Alongside this, the use of deep learning for analyzing thermograms has expanded significantly. However, the limited number of labeled thermograms in existing datasets hinders the effective application of deep neural networks. This paper proposes a novel solution to address the challenge of insufficient and imbalanced data by generating synthetic thermograms through a cyclic mapping approach. Such a strategy incorporates the most distinction with the characteristics of the alternative class, during the procedure of constructing fake thermograms belonging to each of the healthy or cancerous classes. Therefore, the deep model which is formed with these thermograms has a better separation potential between these two oppose types of breast tissues. Applying the proposed method on the thermograms of the DMR-IR dataset indicates a significant improvement in the separation of healthy and cancerous cases compared to methods which utilize standard Generative Adversarial Network (i.e., GAN) for production of fake thermograms. The obtained results demonstrate that the proposed scheme may promote characteristics of specificity and accuracy at least 15 and 10 percent compared to existing methods. In terms of sensitivity, a slight improvement up to 1% have also been observed. Furthermore, the proposed scheme reduces the gap between training and validation performance by about one-third and extends the number of effective training epochs by up to four times, helping to prevent premature overfitting in deep neural network training.