Scholarlink

<p>Brain tumors arising from glioma, meningioma, and pituitary regions present major clinical challenges because of their malignant potential and substantial impact on survival outcomes. Accurate and timely diagnosis remains essential to optimize therapeutic strategies, although most prior studies have concentrated on binary MRI-based classification with limited focus on systematic preprocessing. This work introduces an integrated framework for multi-class brain tumor prediction that incorporates sequential preprocessing steps including resizing, Gaussian blurring, and histogram equalization, followed by a hybrid CNN-Attention-SVM (CASVM) model. Convolutional layers are employed to extract hierarchical features, the attention mechanism selectively emphasizes salient tumor-related regions, and a One versus Rest SVM enables robust multi-class discrimination among glioma, meningioma, pituitary, and healthy cases. The method was evaluated on 7023 MRI scans obtained from Figshare, SARTAJ, and Br35H, achieving 96.86% ± 0.41% accuracy under fivefold cross-validation, thereby surpassing conventional pipelines. These findings highlight the central role of systematic preprocessing and hybrid deep learning SVM integration in achieving accurate, generalizable, and computationally efficient brain tumor diagnostics.</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

Decoding brain tumors: an attention-guided hybrid learning framework for multi-class classification of MRI images

  • 期刊论文

  • 发表日期：

    2025年12月8日

查看全文
  • Soham Bandyopadhyay,
  • Arnab Banerjee,
  • Bhumik Roy,
  • Ankur Lodh

摘要

Brain tumors arising from glioma, meningioma, and pituitary regions present major clinical challenges because of their malignant potential and substantial impact on survival outcomes. Accurate and timely diagnosis remains essential to optimize therapeutic strategies, although most prior studies have concentrated on binary MRI-based classification with limited focus on systematic preprocessing. This work introduces an integrated framework for multi-class brain tumor prediction that incorporates sequential preprocessing steps including resizing, Gaussian blurring, and histogram equalization, followed by a hybrid CNN-Attention-SVM (CASVM) model. Convolutional layers are employed to extract hierarchical features, the attention mechanism selectively emphasizes salient tumor-related regions, and a One versus Rest SVM enables robust multi-class discrimination among glioma, meningioma, pituitary, and healthy cases. The method was evaluated on 7023 MRI scans obtained from Figshare, SARTAJ, and Br35H, achieving 96.86% ± 0.41% accuracy under fivefold cross-validation, thereby surpassing conventional pipelines. These findings highlight the central role of systematic preprocessing and hybrid deep learning SVM integration in achieving accurate, generalizable, and computationally efficient brain tumor diagnostics.