Scholarlink

<p>Prostate cancer (PCa) remains a major global health concern and ranks among the leading causes of cancer-related deaths in men. Magnetic resonance imaging (MRI) provides valuable diagnostic information for PCa detection and treatment planning but poses challenges due to image variability and complex interpretation. To address these issues, this study presents ProAttNet, a systematically integrated residual U-Net architecture with multi-level complementary attention mechanisms for accurate prostate segmentation. Preprocessing with Contrast-Limited Adaptive Histogram Equalization (CLAHE) enhances MRI contrast, while region-based loss functions—Dice Loss (DL) and Intersection over Union (IoU) Loss—mitigate class imbalance and improve delineation accuracy. Evaluations on the PROMISE12 and Medical Segmentation Decathlon (MSD) datasets demonstrate the effectiveness of ProAttNet, achieving mean Dice Similarity Coefficients (DSC) of 95.28% and 91.09%, respectively. Quantitative analyses using the 95% Hausdorff Distance (HD95), Relative Volume Difference (RVD), and Average Surface Distance (ASD) further demonstrate consistent boundary and volume improvements compared with standard U-Net variants. These findings indicate that ProAttNet provides a robust and computationally efficient framework for automated prostate segmentation, supporting more consistent and precise boundary delineation in clinical and research workflows.</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

ProAttNet: a novel network of prostate segmentation with multi-attention residual U-Net using magnetic resonance images

  • 期刊论文

  • 发表日期：

    2026年3月23日

查看全文
  • R. Deiva Nayagam,
  • D. Selvathi,
  • S. Shalini

摘要

Prostate cancer (PCa) remains a major global health concern and ranks among the leading causes of cancer-related deaths in men. Magnetic resonance imaging (MRI) provides valuable diagnostic information for PCa detection and treatment planning but poses challenges due to image variability and complex interpretation. To address these issues, this study presents ProAttNet, a systematically integrated residual U-Net architecture with multi-level complementary attention mechanisms for accurate prostate segmentation. Preprocessing with Contrast-Limited Adaptive Histogram Equalization (CLAHE) enhances MRI contrast, while region-based loss functions—Dice Loss (DL) and Intersection over Union (IoU) Loss—mitigate class imbalance and improve delineation accuracy. Evaluations on the PROMISE12 and Medical Segmentation Decathlon (MSD) datasets demonstrate the effectiveness of ProAttNet, achieving mean Dice Similarity Coefficients (DSC) of 95.28% and 91.09%, respectively. Quantitative analyses using the 95% Hausdorff Distance (HD95), Relative Volume Difference (RVD), and Average Surface Distance (ASD) further demonstrate consistent boundary and volume improvements compared with standard U-Net variants. These findings indicate that ProAttNet provides a robust and computationally efficient framework for automated prostate segmentation, supporting more consistent and precise boundary delineation in clinical and research workflows.