Scholarlink

<p>Brain tumor infection can lead to cognitive impairments and neurological deficits. Its effects on the nervous system can result in sensory disturbances and motor dysfunction. Brain tumor infections can trigger inflammation and swelling, leading to severe headaches and affecting memory and concentration. Due to its harmful impact, brain tumors have emerged as a global health concern, necessitating the development of AI-driven tools to aid medical professionals in the early diagnosis and classification of multiple types of brain tumors. Previous research has predominantly relied on convolutional neural network CNN-based techniques to predict brain tumors. However, these approaches often utilized binary datasets and struggled to extract optimal features for effective model generalization. In our proposed method, we introduce a novel utilization of the manta-ray foraging optimizer-MRFO, specifically in aggregation with a fitness approach, while also incorporating improved residual block techniques within a dense network framework. To address this gap in the literature, we present a novel MRFO-based residual method to optimize the hierarchical feature representation of tumors using DenseNet-169.Initially, we adopt a conventional approach by selecting a base Dense model from a set of pre-trained DenseNet models. Subsequently, we utilize the MRFO strategy to explore optimal hyperparameter settings for the DenseNet-169 model. In the subsequent phase, we extend this approach by integrating an enhanced residual block into the optimized DenseNet-169 architecture, aimed at further boosting performance. Through this methodology, we achieve a notable accuracy of 96.40%, coupled with a robust hierarchical feature representation. This demonstrates the efficacy of our proposed model, surpassing state-of-the-art methods on the same dataset. Furthermore, to validate the robustness of our approach, we conduct experiments on four diverse datasets such as MSID (Multi Skin Infection dataset), gastrointestinal multi-infection, Kaggle MRI multiclass, and BR35H MRI binary class dataset. The proposed model consistently maintains high performance with accuracies of 93.10, 97.46, 98.77, and 99.10% respectively.</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

Optimize brain tumor multiclass classification with manta ray foraging and improved residual block techniques

  • 期刊论文

  • 发表日期：

    2025年2月1日

查看全文
  • Saif Ur Rehman Khan,
  • Sohaib Asif,
  • Ming Zhao,
  • Wei Zou,
  • Yangfan Li

摘要

Brain tumor infection can lead to cognitive impairments and neurological deficits. Its effects on the nervous system can result in sensory disturbances and motor dysfunction. Brain tumor infections can trigger inflammation and swelling, leading to severe headaches and affecting memory and concentration. Due to its harmful impact, brain tumors have emerged as a global health concern, necessitating the development of AI-driven tools to aid medical professionals in the early diagnosis and classification of multiple types of brain tumors. Previous research has predominantly relied on convolutional neural network CNN-based techniques to predict brain tumors. However, these approaches often utilized binary datasets and struggled to extract optimal features for effective model generalization. In our proposed method, we introduce a novel utilization of the manta-ray foraging optimizer-MRFO, specifically in aggregation with a fitness approach, while also incorporating improved residual block techniques within a dense network framework. To address this gap in the literature, we present a novel MRFO-based residual method to optimize the hierarchical feature representation of tumors using DenseNet-169.Initially, we adopt a conventional approach by selecting a base Dense model from a set of pre-trained DenseNet models. Subsequently, we utilize the MRFO strategy to explore optimal hyperparameter settings for the DenseNet-169 model. In the subsequent phase, we extend this approach by integrating an enhanced residual block into the optimized DenseNet-169 architecture, aimed at further boosting performance. Through this methodology, we achieve a notable accuracy of 96.40%, coupled with a robust hierarchical feature representation. This demonstrates the efficacy of our proposed model, surpassing state-of-the-art methods on the same dataset. Furthermore, to validate the robustness of our approach, we conduct experiments on four diverse datasets such as MSID (Multi Skin Infection dataset), gastrointestinal multi-infection, Kaggle MRI multiclass, and BR35H MRI binary class dataset. The proposed model consistently maintains high performance with accuracies of 93.10, 97.46, 98.77, and 99.10% respectively.