Scholarlink

<p>Brain tumors are a leading cause of mortality and morbidity worldwide, necessitating accurate segmentation for effective treatment planning and monitoring. Despite advancements, existing segmentation methods face challenges such as limited annotated datasets, variability in tumor appearance, and lack of model generalizability. In this paper, we propose a robust brain tumor detection and segmentation approach using encoder-decoder architectures, including U-Net, Attention U-Net, ResUnet++, and UNET with Transformers (UNETR). Our methodology employs preprocessing techniques like additive Gaussian noise removal, normalization, and histogram equalization to enhance performance. Additionally, we integrate a fully connected Conditional Random Field (CRF) as a post-processing step to minimize false positives. We utilize HSV color space images for improved segmentation accuracy. We validate our approach on MRI datasets from multiple scanner manufacturers (GE, Siemens, Philips) using metrics such as Dice coefficient, Intersection over Union (IoU), precision, and recall. The results demonstrate superior generalizability, precise segmentation across varied tumor morphologies, and enhanced efficiency in real-world clinical scenarios.</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

Advanced Brain Tumor Detection and Segmentation Using Encoder–Decoder Architectures and Conditional Random Fields

  • 期刊论文

  • 发表日期：

    2026年5月13日

查看全文
  • Ayush Tiwari,
  • Rahul Dixit,
  • Priyank Jain,
  • Urmila Abhijeet Patil

摘要

Brain tumors are a leading cause of mortality and morbidity worldwide, necessitating accurate segmentation for effective treatment planning and monitoring. Despite advancements, existing segmentation methods face challenges such as limited annotated datasets, variability in tumor appearance, and lack of model generalizability. In this paper, we propose a robust brain tumor detection and segmentation approach using encoder-decoder architectures, including U-Net, Attention U-Net, ResUnet++, and UNET with Transformers (UNETR). Our methodology employs preprocessing techniques like additive Gaussian noise removal, normalization, and histogram equalization to enhance performance. Additionally, we integrate a fully connected Conditional Random Field (CRF) as a post-processing step to minimize false positives. We utilize HSV color space images for improved segmentation accuracy. We validate our approach on MRI datasets from multiple scanner manufacturers (GE, Siemens, Philips) using metrics such as Dice coefficient, Intersection over Union (IoU), precision, and recall. The results demonstrate superior generalizability, precise segmentation across varied tumor morphologies, and enhanced efficiency in real-world clinical scenarios.