Scholarlink

<p>Accurate classification of brain tumors from MRI images is crucial for guiding treatment planning and improving clinical outcomes. However, current methods face limitations in generalization due to small datasets and lack of interpretability, both critical in medical applications. To address these challenges, we propose a multi-class brain tumor classification system, exploring three distinct models based on VGG19 architecture. These include an attention-module-based model, a deeper convolutional network, and a linear-boosted model. Each model is trained on a large, aggregated dataset with six types of image augmentations to improve generalization. Among the three models, the attention-based model achieves the highest classification performance by focusing on relevant tumor regions. Local interpretable model-agnostic explanations are used to visualize the decision-making process, enhancing model transparency. Our results demonstrate that the attention-based model outperforms baseline and state-of-the-art methods, making it a robust and interpretable solution for brain tumor diagnosis.</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

Multi-modal MRI data augmentation and attention-based modeling for interpretable brain tumor classification

  • 期刊论文

  • 发表日期：

    2025年8月11日

查看全文
  • Mubarak A. Alanazi

摘要

Accurate classification of brain tumors from MRI images is crucial for guiding treatment planning and improving clinical outcomes. However, current methods face limitations in generalization due to small datasets and lack of interpretability, both critical in medical applications. To address these challenges, we propose a multi-class brain tumor classification system, exploring three distinct models based on VGG19 architecture. These include an attention-module-based model, a deeper convolutional network, and a linear-boosted model. Each model is trained on a large, aggregated dataset with six types of image augmentations to improve generalization. Among the three models, the attention-based model achieves the highest classification performance by focusing on relevant tumor regions. Local interpretable model-agnostic explanations are used to visualize the decision-making process, enhancing model transparency. Our results demonstrate that the attention-based model outperforms baseline and state-of-the-art methods, making it a robust and interpretable solution for brain tumor diagnosis.