Scholarlink

<p>The identification and segmentation of brain tumor anomalies are vital for clinical decision-making; nevertheless, current MRI-based segmentation models continue to face challenges such as elevated feature noise, inadequate generalization, and inconsistent tumor boundary delineation. Despite the application of several deep learning methodologies, including CNN-based and hybrid encoder-decoder networks, their efficacy is often constrained by redundant features, overfitting, and insufficient robustness to intricate MRI datasets. This paper introduces an innovative Red Fox-based ZFNet Encoder (RFZE) architecture that combines evolutionary feature optimization with deep representation learning to enhance BT segmentation accuracy. The model uses a dropout-based noise filtering mechanism and a Red Fox Optimization technique to improve feature selection, and subsequently employs mask-guided segmentation to ensure boundary accuracy. The BRATS dataset was used for evaluation, where the proposed RFZE achieved a segmentation accuracy of 99.8%, surpassing leading models. The results indicate that integrating meta-heuristic optimization into deep encoders markedly enhances the reliability of MRI tumor segmentation, especially in differentiating benign from malignant areas.</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

An optimized deep learning network approach for brain tumor classification of 3D MRI images

  • 期刊论文

  • 发表日期：

    2026年2月6日

查看全文
  • J. Hima Bindu,
  • M. Uma Devi

摘要

The identification and segmentation of brain tumor anomalies are vital for clinical decision-making; nevertheless, current MRI-based segmentation models continue to face challenges such as elevated feature noise, inadequate generalization, and inconsistent tumor boundary delineation. Despite the application of several deep learning methodologies, including CNN-based and hybrid encoder-decoder networks, their efficacy is often constrained by redundant features, overfitting, and insufficient robustness to intricate MRI datasets. This paper introduces an innovative Red Fox-based ZFNet Encoder (RFZE) architecture that combines evolutionary feature optimization with deep representation learning to enhance BT segmentation accuracy. The model uses a dropout-based noise filtering mechanism and a Red Fox Optimization technique to improve feature selection, and subsequently employs mask-guided segmentation to ensure boundary accuracy. The BRATS dataset was used for evaluation, where the proposed RFZE achieved a segmentation accuracy of 99.8%, surpassing leading models. The results indicate that integrating meta-heuristic optimization into deep encoders markedly enhances the reliability of MRI tumor segmentation, especially in differentiating benign from malignant areas.