Scholarlink

In this paper, we focused on a novel Half-UNet CBAM Ghost Model that integrates Convolutional Block Attention Module (CBAM) mechanisms within the Ghost Modules to enhance medical image segmentation. Our proposed architecture leverages the strengths of the traditional U-Net while incorporating CBAM to improve feature representation through channel and spatial attention. The model was evaluated using the MICCAI 2009 Left Ventricle Segmentation Challenge dataset, consisting of MRI images resized to 256 × 256 pixels. Quantitative results demonstrate that the Half-UNet CBAM Ghost Model outperforms both the baseline U-Net and Half-UNet models, achieving a precision of 0.7101, recall of 0.7694, specificity of 0.9949, IoU of 0.5855, and an overall evaluation metric score of 0.7638. Visual comparisons further confirm the segmentation accuracy and detail captured by the proposed model.

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

AHU: Novel Attention Based Half-UNet for Left Ventricle MRI Segmentation

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Surbhi Verma,
  • Hardeo Thakur,
  • Mohit Agarwal

摘要

In this paper, we focused on a novel Half-UNet CBAM Ghost Model that integrates Convolutional Block Attention Module (CBAM) mechanisms within the Ghost Modules to enhance medical image segmentation. Our proposed architecture leverages the strengths of the traditional U-Net while incorporating CBAM to improve feature representation through channel and spatial attention. The model was evaluated using the MICCAI 2009 Left Ventricle Segmentation Challenge dataset, consisting of MRI images resized to 256 × 256 pixels. Quantitative results demonstrate that the Half-UNet CBAM Ghost Model outperforms both the baseline U-Net and Half-UNet models, achieving a precision of 0.7101, recall of 0.7694, specificity of 0.9949, IoU of 0.5855, and an overall evaluation metric score of 0.7638. Visual comparisons further confirm the segmentation accuracy and detail captured by the proposed model.