Scholarlink

<p>Brain Tumor is one of the most devastating disorders due to its high mortality rate and aggressive nature. Recently, the brain tumor classification utilizing Magnetic Resonance Imaging (MRI) gained more attention in medical diagnostics, due to its potential to enhance the early detection for providing subsequent treatment strategies. Despite the fact that the existing methods have achieved better results in brain tumor classification, there are still some difficulties due to variability and time constraints. Hence, the Sonar Prey Optimized Multi-head Convolutional Bidirectional Memory (SPO-MCBM) model is proposed to classify the brain tumors in the early stage using the MRI images. Further, the proposed SPO-MCBM method exploits the Hybrid Structural Statistical Optical flow Matrix-based Features (HS<sup>2</sup>OM), contributing to minimizing the computational complexity and providing high classification accuracy. The incorporation of an Optimized Fused Attention-based W-Net (OFA-W-Net) model for segmentation minimizes the processing time of classification by accurately segmenting the tumor region. Additionally, the incorporation of the Multi-head Attention (MA) improves the model’s learning process by selectively focusing on different critical regions in the input sequence, capturing the complex relationships and minimize the redundancy in the features. Specifically, the application of the Sonar Prey Optimization (SPO) optimally tunes the SPO-MCBM model, leading to high classification accuracy. Extensive experiments reveal that the proposed SPO-MCBM model reported the high accuracy of 97.92%, precision of 96.71%, <i>F</i>1-score of 97.61%, recall of 98.52%, FNR of 0.014, and FPR of 0.032 for 90% of training utilizing the Brain tumor dataset. Furthermore, the proposed method exhibits the best performance with achieving the high accuracy, <i>F</i>1-score, precision, recall, FNR, and FPR of 96.53%, 96.62%, 95.49%, 97.77%, 0.022, and 0.045 for BraTS 2020 dataset, indicating the model’s superiority over other state-of-the-art models for brain tumor classification.</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

SPO-MCBM: Sonar Prey Optimized Multi-head Convolutional Bidirectional Memory Network for Brain Tumor Classification Using MRI

  • 期刊论文

  • 发表日期：

    2025年10月21日

查看全文
  • Aasha Mahesh Chavan,
  • Vanita Mane

摘要

Brain Tumor is one of the most devastating disorders due to its high mortality rate and aggressive nature. Recently, the brain tumor classification utilizing Magnetic Resonance Imaging (MRI) gained more attention in medical diagnostics, due to its potential to enhance the early detection for providing subsequent treatment strategies. Despite the fact that the existing methods have achieved better results in brain tumor classification, there are still some difficulties due to variability and time constraints. Hence, the Sonar Prey Optimized Multi-head Convolutional Bidirectional Memory (SPO-MCBM) model is proposed to classify the brain tumors in the early stage using the MRI images. Further, the proposed SPO-MCBM method exploits the Hybrid Structural Statistical Optical flow Matrix-based Features (HS2OM), contributing to minimizing the computational complexity and providing high classification accuracy. The incorporation of an Optimized Fused Attention-based W-Net (OFA-W-Net) model for segmentation minimizes the processing time of classification by accurately segmenting the tumor region. Additionally, the incorporation of the Multi-head Attention (MA) improves the model’s learning process by selectively focusing on different critical regions in the input sequence, capturing the complex relationships and minimize the redundancy in the features. Specifically, the application of the Sonar Prey Optimization (SPO) optimally tunes the SPO-MCBM model, leading to high classification accuracy. Extensive experiments reveal that the proposed SPO-MCBM model reported the high accuracy of 97.92%, precision of 96.71%, F1-score of 97.61%, recall of 98.52%, FNR of 0.014, and FPR of 0.032 for 90% of training utilizing the Brain tumor dataset. Furthermore, the proposed method exhibits the best performance with achieving the high accuracy, F1-score, precision, recall, FNR, and FPR of 96.53%, 96.62%, 95.49%, 97.77%, 0.022, and 0.045 for BraTS 2020 dataset, indicating the model’s superiority over other state-of-the-art models for brain tumor classification.