Scholarlink

<p>In recent years, deep learning algorithms have been extensively examined as tools to assist ultrasound physicians by reducing diagnostic challenges. During practical ultrasound examinations, doctors frequently observe dynamic ultrasound video sequences. In contrast to static ultrasound scans, dynamic imaging offers enhanced information regarding malignancies, facilitating more precise diagnoses. In tumor research, ultrasound professionals employ two fundamental modes: one emphasizes specific areas in static ultrasound images, while the other analyzes temporal connections between successive frames in ultrasound videos. The backbone of MambaCA-Net has two components: the Single-frame Image Stream (SIS) for static ultrasound characteristics and the Continuous-frame Video Stream (CVS) for dynamic ultrasound characteristics. Our model was evaluated using a breast ultrasound dataset including 248 patients and 35,642 images. The backbone of MambaCA-Net consists of two components: the Single-frame Image Stream (SIS) for static ultrasound characteristics and the Continuous-frame Video Stream (CVS) for dynamic ultrasound characteristics. We evaluated our model using a breast ultrasound dataset including 248 patients and 35,642 images. The findings indicate that MambaCA-Net achieves an accuracy of 86.21% and an AUC of 93%. Moreover, we demonstrate that including the CVS enhanced model accuracy by 1.84%. Our data indicate that MambaCA-Net shows significant potential for breast cancer detection. Its lightweight architecture and robust feature extraction capabilities render it optimal for breast cancer diagnostic tasks, even in environments with limited computational resources.</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

MambaCA-Net: a hybrid dual-stream architecture integrating temporal and spatial features for breast ultrasound analysis

  • 期刊论文

  • 发表日期：

    2025年9月22日

查看全文
  • Xu He,
  • Yu Yan,
  • Man Wang,
  • Runqiu Cai,
  • Jingwu Ma,
  • Xiaowei Cai,
  • Ying Tong

摘要

In recent years, deep learning algorithms have been extensively examined as tools to assist ultrasound physicians by reducing diagnostic challenges. During practical ultrasound examinations, doctors frequently observe dynamic ultrasound video sequences. In contrast to static ultrasound scans, dynamic imaging offers enhanced information regarding malignancies, facilitating more precise diagnoses. In tumor research, ultrasound professionals employ two fundamental modes: one emphasizes specific areas in static ultrasound images, while the other analyzes temporal connections between successive frames in ultrasound videos. The backbone of MambaCA-Net has two components: the Single-frame Image Stream (SIS) for static ultrasound characteristics and the Continuous-frame Video Stream (CVS) for dynamic ultrasound characteristics. Our model was evaluated using a breast ultrasound dataset including 248 patients and 35,642 images. The backbone of MambaCA-Net consists of two components: the Single-frame Image Stream (SIS) for static ultrasound characteristics and the Continuous-frame Video Stream (CVS) for dynamic ultrasound characteristics. We evaluated our model using a breast ultrasound dataset including 248 patients and 35,642 images. The findings indicate that MambaCA-Net achieves an accuracy of 86.21% and an AUC of 93%. Moreover, we demonstrate that including the CVS enhanced model accuracy by 1.84%. Our data indicate that MambaCA-Net shows significant potential for breast cancer detection. Its lightweight architecture and robust feature extraction capabilities render it optimal for breast cancer diagnostic tasks, even in environments with limited computational resources.