Scholarlink

<p>In the field of magnetic resonance imaging (MRI), the automatic segmentation of brain tumors is crucial for accurate diagnoses and effective treatment strategies. The precise identification of tumor regions is fundamental in assessing patient health, driving continuous innovations in Deep Learning techniques for auto- mated segmentation. This study presents a novel and efficient approach based on the Res-UNet model for three-dimensional segmentation of multimodal MRI scans. The proposed 3D AGRes-UNet model incorporates a multi-scale feature extraction module (ASPP) to capture information at diverse spatial resolutions and utilizes Attention Gates mechanisms highlight specific regions of interest within the image. Evaluated on the BraTS 2020 dataset, the model demon- strated promising results, achieving Dice coefficients of 0.8059, 0.7621, and 0.7224 for whole tumor (WT), tumor core (TC), and enhancing tumor (ET), respectively. A comparative evaluation with state-of-the-art models reveals the superior segmentation performance of 3D AGRes-UNet, as evidenced by different metrics.</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

3D AGRes-UNet: an advanced deep learning model for brain tumor segmentation from multimodal MRI

  • 期刊论文

  • 发表日期：

    2025年4月12日

查看全文
  • Khaoula Echine,
  • Aziz Darouichi

摘要

In the field of magnetic resonance imaging (MRI), the automatic segmentation of brain tumors is crucial for accurate diagnoses and effective treatment strategies. The precise identification of tumor regions is fundamental in assessing patient health, driving continuous innovations in Deep Learning techniques for auto- mated segmentation. This study presents a novel and efficient approach based on the Res-UNet model for three-dimensional segmentation of multimodal MRI scans. The proposed 3D AGRes-UNet model incorporates a multi-scale feature extraction module (ASPP) to capture information at diverse spatial resolutions and utilizes Attention Gates mechanisms highlight specific regions of interest within the image. Evaluated on the BraTS 2020 dataset, the model demon- strated promising results, achieving Dice coefficients of 0.8059, 0.7621, and 0.7224 for whole tumor (WT), tumor core (TC), and enhancing tumor (ET), respectively. A comparative evaluation with state-of-the-art models reveals the superior segmentation performance of 3D AGRes-UNet, as evidenced by different metrics.