Scholarlink

<p>To ensure timely prediction and halt the advancement of brain tumor diseases, automating brain tumor detection is imperative within the medical realm. Despite advancements, the current methods for detecting and classifying brain tumors are hindered by their limited accuracy and higher computation cost. In our study, we present a novel model that surpasses deep learning approaches and existing studies in both performance and complexity. In order to enhance the combination of backbone features into the residual block, we introduce three additional layers. These layers are meticulously made to synergize with the output of the lighter and simpler backbone architecture. Their key role is to smoothly relay feature from the backbone network to the residual block while maintaining all relevant information generated by the model intact. We have upgraded our architecture with a Residual block featuring a separable convolution layer alongside conventional convolutional layers. Replacing ReLU with Swish enhances information retention and broadens feature generation. We utilized two publicly available benchmark datasets, namely Figshare Multiclass and BR35H Binary Class, consisting of 3063 and 3000 sample images, respectively, to assess the efficacy and adaptability of the proposed method. Our model achieves an accuracy of 99.83% on the BR35H binary dataset and 96.95% on the Figshare multiclass dataset, outperforming most pre-trained models on benchmark datasets related to brain tumors while using fewer parameters and computations, showcasing its computational efficiency. This approach holds promise for tumor surveillance in communities with limited healthcare resources, with plans to expand the dataset for broader applicability.</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

Detection of MRI brain tumor using residual skip block based modified MobileNet model

  • 期刊论文

  • 发表日期：

    2025年2月25日

查看全文
  • Saif Ur Rehman Khan,
  • Ming Zhao,
  • Yangfan Li

摘要

To ensure timely prediction and halt the advancement of brain tumor diseases, automating brain tumor detection is imperative within the medical realm. Despite advancements, the current methods for detecting and classifying brain tumors are hindered by their limited accuracy and higher computation cost. In our study, we present a novel model that surpasses deep learning approaches and existing studies in both performance and complexity. In order to enhance the combination of backbone features into the residual block, we introduce three additional layers. These layers are meticulously made to synergize with the output of the lighter and simpler backbone architecture. Their key role is to smoothly relay feature from the backbone network to the residual block while maintaining all relevant information generated by the model intact. We have upgraded our architecture with a Residual block featuring a separable convolution layer alongside conventional convolutional layers. Replacing ReLU with Swish enhances information retention and broadens feature generation. We utilized two publicly available benchmark datasets, namely Figshare Multiclass and BR35H Binary Class, consisting of 3063 and 3000 sample images, respectively, to assess the efficacy and adaptability of the proposed method. Our model achieves an accuracy of 99.83% on the BR35H binary dataset and 96.95% on the Figshare multiclass dataset, outperforming most pre-trained models on benchmark datasets related to brain tumors while using fewer parameters and computations, showcasing its computational efficiency. This approach holds promise for tumor surveillance in communities with limited healthcare resources, with plans to expand the dataset for broader applicability.