Scholarlink

<p>Breast cancer is one of the leading causes of death for women globally. To increase the survival rates associated with these therapies and results, early and accurate diagnosis of the tumor and subsequent appropriate tumor staging are the only options. To overcome this challenge, this research proposes a Random Graph Diffusion Two-Branch Attention Adversarial Domain Adaptation Network with an Improved Orca Predation Algorithm (RGDTAADA-IOPA). Initially, ultrasound (US) and mammography (MG) images are collected from the BUSI and MIAS datasets for analysis. Following data collection, the Shape-Aware Mesh Normal Filtering (SAMNF) is used as a preprocessing step. This method efficiently reduces noise while maintaining significant geometric patterns necessary for precise analysis by improving normal vectors based on regional surface properties. Following preprocessing, the Anatomy-Aware Hover-Transformer (AAHT) segments tumors precisely. AAHT improves tumor border delineation across all imaging types by capturing spatial and anatomical linkages through the use of a visual transformer architecture augmented with both vertical and horizontal strip embeddings. The Random Graph Diffusion Two-Branch Attention Adversarial Domain Adaptation Network (RGDTAADA) is then employed. In the multi-modal diagnostic context, this unified approach enables cross-domain alignment and dependable learning by combining adversarial domain adaptation with spatial–temporal attention. Lastly, the enhanced orca predation algorithm (IOPA), which uses naturally inspired search behaviors to optimize the RGDTAADA hyperparameter and repeatedly search and exploit superior parameter space, has improved model performance and robustness in diagnosing breast cancer. This 99.98 accuracy rate, 99.96 precision rate, 99.93 F1—rate, and 99.95 specificity rate that is achieved by the proposed approach is indeed a highly reliable tool in the medical industry since it will be highly accurate when diagnosing breast cancer.</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

Enhanced Multimodal Breast Cancer Diagnosis Using Random Graph Diffusion-Based Two-Branch Attention Adversarial Domain Adaptation with Improved Orca Predation Optimization

  • 期刊论文

  • 发表日期：

    2025年8月4日

查看全文
  • D. Vetrithangam,
  • Dinesh Kumar Anguraj,
  • Krishna Prakash Arunachalam,
  • Nidhya Rangarajan

摘要

Breast cancer is one of the leading causes of death for women globally. To increase the survival rates associated with these therapies and results, early and accurate diagnosis of the tumor and subsequent appropriate tumor staging are the only options. To overcome this challenge, this research proposes a Random Graph Diffusion Two-Branch Attention Adversarial Domain Adaptation Network with an Improved Orca Predation Algorithm (RGDTAADA-IOPA). Initially, ultrasound (US) and mammography (MG) images are collected from the BUSI and MIAS datasets for analysis. Following data collection, the Shape-Aware Mesh Normal Filtering (SAMNF) is used as a preprocessing step. This method efficiently reduces noise while maintaining significant geometric patterns necessary for precise analysis by improving normal vectors based on regional surface properties. Following preprocessing, the Anatomy-Aware Hover-Transformer (AAHT) segments tumors precisely. AAHT improves tumor border delineation across all imaging types by capturing spatial and anatomical linkages through the use of a visual transformer architecture augmented with both vertical and horizontal strip embeddings. The Random Graph Diffusion Two-Branch Attention Adversarial Domain Adaptation Network (RGDTAADA) is then employed. In the multi-modal diagnostic context, this unified approach enables cross-domain alignment and dependable learning by combining adversarial domain adaptation with spatial–temporal attention. Lastly, the enhanced orca predation algorithm (IOPA), which uses naturally inspired search behaviors to optimize the RGDTAADA hyperparameter and repeatedly search and exploit superior parameter space, has improved model performance and robustness in diagnosing breast cancer. This 99.98 accuracy rate, 99.96 precision rate, 99.93 F1—rate, and 99.95 specificity rate that is achieved by the proposed approach is indeed a highly reliable tool in the medical industry since it will be highly accurate when diagnosing breast cancer.