Scholarlink

<b>Purpose:</b> <p>Mammography is the most commonly used method for early screening of breast cancer. It is of great research value to develop a computer-aided diagnosis (CAD) system based on mammography images to assist doctors in making efficient and accurate diagnoses. Mass segmentation in mammograms is a core component of breast cancer CAD systems and an essential step in further qualitative analysis of breast cancer. However, existing methods still have difficulty in capturing multi-scale mass in complex backgrounds, especially when the mass is occluded by tissues with similar appearance.</p> <b>Methods:</b> <p>To address these challenges, this paper proposes a context-aware feature complementary screening network for mass segmentation in whole mammograms (CAF-CS), which achieves accurate breast mass segmentation through multi-level feature fusion and screening mechanism. Specifically, the feature complementary module (FCM) adopts a weighted fusion strategy to complement low-level detail features (edge and texture) with high-level semantic features (lesion type), effectively solving the problem of context information loss in traditional methods. The feature screening module (FSM) adopts a dual attention mechanism to strengthen global semantic information in the channel dimension and focus on local key areas in the spatial dimension, thereby screening out the most discriminative lesion features.</p> <b>Results:</b> <p>Evaluation results of CAF-CS on INbreast, CBIS-DDSM and private In-house datasets show that it can generate more accurate mass masks in complex scenarios.</p> <b>Conclusion:</b> <p>CAF-CS is dedicated to providing strong support for the application of breast cancer CAD systems.</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

Context-aware feature complementary screening network for mass segmentation in whole mammograms

  • 期刊论文

  • 发表日期：

    2025年12月4日

查看全文
  • Qingkun Guo,
  • Mei Liu,
  • Luhao Sun,
  • Chao Li,
  • Wenzong Jiang,
  • Weifeng Liu,
  • Lin Cong,
  • Zhiyong Yu,
  • Baodi Liu

摘要

Purpose:

Mammography is the most commonly used method for early screening of breast cancer. It is of great research value to develop a computer-aided diagnosis (CAD) system based on mammography images to assist doctors in making efficient and accurate diagnoses. Mass segmentation in mammograms is a core component of breast cancer CAD systems and an essential step in further qualitative analysis of breast cancer. However, existing methods still have difficulty in capturing multi-scale mass in complex backgrounds, especially when the mass is occluded by tissues with similar appearance.

Methods:

To address these challenges, this paper proposes a context-aware feature complementary screening network for mass segmentation in whole mammograms (CAF-CS), which achieves accurate breast mass segmentation through multi-level feature fusion and screening mechanism. Specifically, the feature complementary module (FCM) adopts a weighted fusion strategy to complement low-level detail features (edge and texture) with high-level semantic features (lesion type), effectively solving the problem of context information loss in traditional methods. The feature screening module (FSM) adopts a dual attention mechanism to strengthen global semantic information in the channel dimension and focus on local key areas in the spatial dimension, thereby screening out the most discriminative lesion features.

Results:

Evaluation results of CAF-CS on INbreast, CBIS-DDSM and private In-house datasets show that it can generate more accurate mass masks in complex scenarios.

Conclusion:

CAF-CS is dedicated to providing strong support for the application of breast cancer CAD systems.