Scholarlink

<p>Brain tumors from MRI scans must be classified quickly and precisely to increase diagnosis accuracy and patient outcomes. In this study, we presented a novel deep learning (DL) architecture built upon a fine-tuned Xception architecture, enhanced through the sequential integration of two attention modules: Convolutional Block Attention Module (CBAM) &amp; Squeeze-and-Excitation (SE). This dual-stage attention mechanism enables progressive refinement of feature representations by first amplifying informative channels and then localizing critical spatial regions, thereby improving the model’s focus on tumor-specific patterns. A customized transfer learning strategy was applied to adapt the network to domain-specific medical data, addressing the limitations of data scarcity and domain shift. The suggested model was trained and evaluated on the Magnetic Resonance Imaging (MRI) Dataset (version 1.0) from Kaggle, including the following four classes: pituitary tumour, meningioma, glioma, and no tumour. The proposed model demonstrates strong generalization and robustness by achieving a classification accuracy of 99.47%. While the primary contribution lies in the architectural innovation, empirical results also reflect consistent performance improvements in assessment measures, which include accuracy, AUC, F1-score, recall, and precision. The framework is computationally efficient and clinically scalable, offering a promising solution for reliable, automated brain tumor diagnosis in real-world neuroimaging applications.</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 Brain Tumor Detection in MRI Scans Using Deep Learning Approaches

  • 期刊论文

  • 发表日期：

    2025年10月8日

查看全文
  • Jyoti,
  • Anuj Kumar,
  • Silky Sachar

摘要

Brain tumors from MRI scans must be classified quickly and precisely to increase diagnosis accuracy and patient outcomes. In this study, we presented a novel deep learning (DL) architecture built upon a fine-tuned Xception architecture, enhanced through the sequential integration of two attention modules: Convolutional Block Attention Module (CBAM) & Squeeze-and-Excitation (SE). This dual-stage attention mechanism enables progressive refinement of feature representations by first amplifying informative channels and then localizing critical spatial regions, thereby improving the model’s focus on tumor-specific patterns. A customized transfer learning strategy was applied to adapt the network to domain-specific medical data, addressing the limitations of data scarcity and domain shift. The suggested model was trained and evaluated on the Magnetic Resonance Imaging (MRI) Dataset (version 1.0) from Kaggle, including the following four classes: pituitary tumour, meningioma, glioma, and no tumour. The proposed model demonstrates strong generalization and robustness by achieving a classification accuracy of 99.47%. While the primary contribution lies in the architectural innovation, empirical results also reflect consistent performance improvements in assessment measures, which include accuracy, AUC, F1-score, recall, and precision. The framework is computationally efficient and clinically scalable, offering a promising solution for reliable, automated brain tumor diagnosis in real-world neuroimaging applications.