Scholarlink

<p>Early and precise detection of brain tumors from magnetic resonance imaging (MRI) images is critical for improving the patient’s life span. However, manual segmentation and other traditional methods, such as handcrafted feature engineering, often struggle with tumor characteristics’ variability and complexity. Therefore, this study introduces a novel dual-stream neural network, DBTNet, to address these challenges and enhance brain tumor detection from MRI images. The presented DBTNet utilizes two parallel streams: one focuses on capturing essential spatial (or local) features, while the other extracts global (or edge) features from MRI images. In addition, MRI images are also preprocessed to optimize computational efficiency. This dual-stream network enables comprehensive feature representation, facilitating accurate tumor classification. The proposed system is evaluated on two publicly available datasets from Kaggle: a binary and a multi-class dataset. It achieved an accuracy of 99.58% in binary classification and 97.90% in multi-class classification tasks, which is superior compared to state-of-the-art methods. These results show the efficiency of the suggested dual-stream approach in precisely identifying tumors across diverse MRI datasets.</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

DBTNet: A Dual-Stream Neural Network for Effective Brain Tumor Detection in MRI Images

  • 期刊论文

  • 发表日期：

    2025年5月29日

查看全文
  • Rasool Reddy Kamireddy,
  • Vijayakumar Kadha,
  • Kandala NVPS Rajesh,
  • Ravindra Dhuli,
  • Sadiq Hussain

摘要

Early and precise detection of brain tumors from magnetic resonance imaging (MRI) images is critical for improving the patient’s life span. However, manual segmentation and other traditional methods, such as handcrafted feature engineering, often struggle with tumor characteristics’ variability and complexity. Therefore, this study introduces a novel dual-stream neural network, DBTNet, to address these challenges and enhance brain tumor detection from MRI images. The presented DBTNet utilizes two parallel streams: one focuses on capturing essential spatial (or local) features, while the other extracts global (or edge) features from MRI images. In addition, MRI images are also preprocessed to optimize computational efficiency. This dual-stream network enables comprehensive feature representation, facilitating accurate tumor classification. The proposed system is evaluated on two publicly available datasets from Kaggle: a binary and a multi-class dataset. It achieved an accuracy of 99.58% in binary classification and 97.90% in multi-class classification tasks, which is superior compared to state-of-the-art methods. These results show the efficiency of the suggested dual-stream approach in precisely identifying tumors across diverse MRI datasets.