Scholarlink

<p>Deep learning models for semantic segmentation, including U-Net, DeepLabV3, and PSPNet, have demonstrated promising results in medical imaging by identifying and delineating disease areas. In these models, convolutional neural networks (CNNs) are employed to extract high-level feature representations from input images, capturing crucial spatial and contextual details necessary for accurate segmentation. To accurately segment breast cancer, CNNs are trained on extensive datasets of labeled ultrasound images. However, conventional segmentation methods often face challenges with low accuracy, particularly for small or variably sized objects, such as tumors or lesions, due to information loss in affected regions. Thus, the proposed approach utilizes U-Net architecture with EfficientNetB7 as the encoder-decoder backbone, enhanced with atrous convolutions for improved feature extraction. To further refine the network, the triplet attention mechanism is introduced as a bottleneck module, and the channel attention module is applied through skip connections to emphasize key spatial and channel-specific features. Prior to segmentation, medical images undergo preprocessing using adaptive dual Gamma correction to enhance image contrast and reduce artifacts. Additionally, the ground matrix is adaptively learned using reinforcement learning, allowing the model to improve its segmentation accuracy over time. From the experimental outcome, it is found that the proposed model provides high accuracy among the existing methods compared with them. The proposed model’s performance is evaluated using several metrics, including intersection over union, precision, recall, dice score, and pixel accuracy. Experimental results indicate that this model significantly outperforms existing methods, demonstrating its effectiveness in accurately segmenting breast cancer in ultrasound images despite the presence of high artifacts. By addressing the challenges associated with ultrasound imaging, the proposed approach aims to enhance the diagnostic accuracy and reliability of breast cancer detection, ultimately contributing to better patient outcomes.</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

Ultrasound Breast Cancer Segmentation Using Atrous Convolution EfficientNetB7 with Attention Module in U-Net

  • 期刊论文

  • 发表日期：

    2025年9月4日

查看全文
  • Suja Paulose,
  • Veera Vanitha Duraisamy

摘要

Deep learning models for semantic segmentation, including U-Net, DeepLabV3, and PSPNet, have demonstrated promising results in medical imaging by identifying and delineating disease areas. In these models, convolutional neural networks (CNNs) are employed to extract high-level feature representations from input images, capturing crucial spatial and contextual details necessary for accurate segmentation. To accurately segment breast cancer, CNNs are trained on extensive datasets of labeled ultrasound images. However, conventional segmentation methods often face challenges with low accuracy, particularly for small or variably sized objects, such as tumors or lesions, due to information loss in affected regions. Thus, the proposed approach utilizes U-Net architecture with EfficientNetB7 as the encoder-decoder backbone, enhanced with atrous convolutions for improved feature extraction. To further refine the network, the triplet attention mechanism is introduced as a bottleneck module, and the channel attention module is applied through skip connections to emphasize key spatial and channel-specific features. Prior to segmentation, medical images undergo preprocessing using adaptive dual Gamma correction to enhance image contrast and reduce artifacts. Additionally, the ground matrix is adaptively learned using reinforcement learning, allowing the model to improve its segmentation accuracy over time. From the experimental outcome, it is found that the proposed model provides high accuracy among the existing methods compared with them. The proposed model’s performance is evaluated using several metrics, including intersection over union, precision, recall, dice score, and pixel accuracy. Experimental results indicate that this model significantly outperforms existing methods, demonstrating its effectiveness in accurately segmenting breast cancer in ultrasound images despite the presence of high artifacts. By addressing the challenges associated with ultrasound imaging, the proposed approach aims to enhance the diagnostic accuracy and reliability of breast cancer detection, ultimately contributing to better patient outcomes.