Scholarlink

<p>Brain tumors represent some of the most critical neurological disorders, where early detection and precise classification are crucial for improving patient outcomes. Magnetic Resonance Imaging (MRI) is a crucial diagnostic modality for identifying and evaluating such tumors. However, the traditional manual interpretation of MRI scans is often time-consuming and prone to inconsistency. To address these challenges, this study proposes an automated framework for MRI-based brain tumor segmentation and classification using U-Net and ReXNet-150 models. The U-Net architecture is utilized to accurately delineate tumor regions, while ReXNet-150 classifies the segmented regions into meningioma, glioma, or no tumor categories. The proposed end-to-end system achieved robust and statistically validated performance, with 96.7% accuracy, precision of 0.952, recall of 0.948, and an F1-score of 0.950, supported by 95% confidence intervals obtained through patient-wise cross-validation. Additionally, the oracle (ground-truth ROI) evaluation achieved 99.3% accuracy, representing the theoretical upper bound of classifier performance. These findings demonstrate that the integrated U-Net → ReXNet-150 pipeline provides a reliable and computationally efficient solution for real-world clinical applications, with a specific focus on meningioma and glioma detection, thereby maintaining anatomical consistency and modeling precision.</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

Automated MRI brain abnormality detection using U-Net and ReX-Net

  • 期刊论文

  • 发表日期：

    2026年6月2日

查看全文
  • Krishna Prakash,
  • S. Sreedhar Babu,
  • Kanumuri Harini,
  • P. Ravi Teja,
  • K. Kiranmai,
  • Shonak Bansal,
  • Mohammad Rashed Iqbal Faruque,
  • K. S. Al-mugren

摘要

Brain tumors represent some of the most critical neurological disorders, where early detection and precise classification are crucial for improving patient outcomes. Magnetic Resonance Imaging (MRI) is a crucial diagnostic modality for identifying and evaluating such tumors. However, the traditional manual interpretation of MRI scans is often time-consuming and prone to inconsistency. To address these challenges, this study proposes an automated framework for MRI-based brain tumor segmentation and classification using U-Net and ReXNet-150 models. The U-Net architecture is utilized to accurately delineate tumor regions, while ReXNet-150 classifies the segmented regions into meningioma, glioma, or no tumor categories. The proposed end-to-end system achieved robust and statistically validated performance, with 96.7% accuracy, precision of 0.952, recall of 0.948, and an F1-score of 0.950, supported by 95% confidence intervals obtained through patient-wise cross-validation. Additionally, the oracle (ground-truth ROI) evaluation achieved 99.3% accuracy, representing the theoretical upper bound of classifier performance. These findings demonstrate that the integrated U-Net → ReXNet-150 pipeline provides a reliable and computationally efficient solution for real-world clinical applications, with a specific focus on meningioma and glioma detection, thereby maintaining anatomical consistency and modeling precision.