Scholarlink

<p>Accurate diagnosis of brain tumors from MRI images is vital for effective treatment planning and improved patient outcomes. However, the manual interpretation of MRI scans remains challenging due to tumor heterogeneity. This paper introduces a block-based Convolutional Neural Network (CNN) model with adaptive channel attention and a multi-path convolutional architecture to enhance tumor detection and classification. The proposed model leverages Efficient Channel Attention (ECA) and convolutional attention mechanisms to capture spatial and channel-wise dependencies, effectively highlighting tumorous regions while suppressing irrelevant information. Evaluated on the Brain Tumor MRI dataset, the model achieved 98.47% accuracy, 99.51% mean average precision (mAP), and an average inference time of 17.2 ms. Comparative analysis with state-of-the-art models-including VGG16, VGG19, ResNet50, MobileNetV3, ShuffleNetV2, and generative methods (Autoencoders, AGAN, and MLGAN) demonstrates that the proposed model outperforms these frameworks in both accuracy and speed, making it well-suited for real-time clinical 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

Adaptive channel attention and multi-path convolutional architecture for brain tumor detection using MRI images

  • 期刊论文

  • 发表日期：

    2025年5月19日

查看全文
  • Muneeb A. Khan,
  • Heemin Park

摘要

Accurate diagnosis of brain tumors from MRI images is vital for effective treatment planning and improved patient outcomes. However, the manual interpretation of MRI scans remains challenging due to tumor heterogeneity. This paper introduces a block-based Convolutional Neural Network (CNN) model with adaptive channel attention and a multi-path convolutional architecture to enhance tumor detection and classification. The proposed model leverages Efficient Channel Attention (ECA) and convolutional attention mechanisms to capture spatial and channel-wise dependencies, effectively highlighting tumorous regions while suppressing irrelevant information. Evaluated on the Brain Tumor MRI dataset, the model achieved 98.47% accuracy, 99.51% mean average precision (mAP), and an average inference time of 17.2 ms. Comparative analysis with state-of-the-art models-including VGG16, VGG19, ResNet50, MobileNetV3, ShuffleNetV2, and generative methods (Autoencoders, AGAN, and MLGAN) demonstrates that the proposed model outperforms these frameworks in both accuracy and speed, making it well-suited for real-time clinical applications.