Scholarlink

<p>Brain tumor detection is a challenge because of the fuzzy growth. The irregular boundaries of tumors lead to inaccurate segmentation of brain tumors. Therefore, to overcome such challenges, a generic segmentation method is proposed named as Javeria Amin segmentation method (JASM) which consists of two phases. The U-network segmentation model is proposed using selected hyperparameters 16 batch-size, 1e-4 learning rate, and 200 epochs for core tumor segmentation. In the&#xa0;second phase, the framework is designed based on the regional proposal network (RPN) model used as the&#xa0;backbone of the pre-trained ResNet-50 model for segmenting core and edema regions. The selected hyperparameters are used to train the framework. Both models are trained on different views of axial, coronal, and sagittal, which help to detect severe&#xa0;brain tumors more accurately. This work is evaluated on AJDBS-2023, Figshare Brain Tumor, BRTAS-2020, and BRTAS-2021 publicly available segmentation datasets. The proposed method provides an&#xa0;Average Dice Score of 0.92 ± 0.015 on AJDBS-2023, 0.92 ± 0.005 on Figshare Brain Tumor, 0.91 ± 0.09 on BRTAS-2020, 0.91 ± 0.004 on BRTAS-2021 datasets. The results prove that the segmentation method can detect the small core tumor region accurately. This method is also partially tested on real patient data in the Hospital under the supervision of expert radiologists that authenticate the contribution of this work based on good performance.</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

Dual-method for semantic and instance brain tumor segmentation based on UNet and mask R-CNN using MRI

  • 期刊论文

  • 发表日期：

    2025年2月7日

查看全文
  • Javaria Amin,
  • Nadia Gul,
  • Muhammad Sharif

摘要

Brain tumor detection is a challenge because of the fuzzy growth. The irregular boundaries of tumors lead to inaccurate segmentation of brain tumors. Therefore, to overcome such challenges, a generic segmentation method is proposed named as Javeria Amin segmentation method (JASM) which consists of two phases. The U-network segmentation model is proposed using selected hyperparameters 16 batch-size, 1e-4 learning rate, and 200 epochs for core tumor segmentation. In the second phase, the framework is designed based on the regional proposal network (RPN) model used as the backbone of the pre-trained ResNet-50 model for segmenting core and edema regions. The selected hyperparameters are used to train the framework. Both models are trained on different views of axial, coronal, and sagittal, which help to detect severe brain tumors more accurately. This work is evaluated on AJDBS-2023, Figshare Brain Tumor, BRTAS-2020, and BRTAS-2021 publicly available segmentation datasets. The proposed method provides an Average Dice Score of 0.92 ± 0.015 on AJDBS-2023, 0.92 ± 0.005 on Figshare Brain Tumor, 0.91 ± 0.09 on BRTAS-2020, 0.91 ± 0.004 on BRTAS-2021 datasets. The results prove that the segmentation method can detect the small core tumor region accurately. This method is also partially tested on real patient data in the Hospital under the supervision of expert radiologists that authenticate the contribution of this work based on good performance.