Scholarlink

<p>Accurate detection of brain tumors in medical images remains a significant challenge, primarily due to the limited diagnostic capabilities of single-modality imaging, which often fails to capture the structural and textural complexity of brain tissues. To overcome these challenges, this paper introduces MultiRegFuse, a robust dual-step framework for multimodal medical image registration and fusion. The proposed method enhances both visual quality and diagnostic precision by integrating feature-based registration with fusion mechanism. In the registration phase, features are extracted using speeded-up robust features (SURF) and refined through alpha trimmed spatial relation correspondence (ATSRC), enabling precise alignment of reference and target images through homography transformation. Following registration, fusion stage employs lifting wavelet transform (LWT) to decompose the images into approximation and detail coefficients. A bidirectional two-stage fusion strategy is applied: F1-fusion merges approximation coefficient of reference image with detail coefficient of register image, while F2-fusion merges approximation coefficient of register image with detail coefficient of reference image, capturing complementary structural and textural information. To determine the optimal contribution from F1-fusion and F2-fusion, united equilibrium optimizer (UEO) algorithm is employed to compute the optimal fusion weights that enhance image quality. The fused coefficients generated through these weighted combinations are then passed through inverse LWT to reconstruct the final fused image. Experimental validation on both standard and clinical multimodal datasets demonstrates that MultiRegFuse consistently outperforms existing techniques across evaluation metrics such as SSIM increased from 0.7502 to 0.92873, PSNR increased from 18.8808 to 41.7691, Entropy increased from 7.423 to 12.4829, Correlation Coefficient improved from 0.9302 to 0.98621, and Mutual Information improved from 5.7087 to 9.45701 confirming its efficacy and reliability for clinical applications in brain tumor diagnosis.</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

MultiRegFuse: Multimodal Medical Image Registration Using SURF-ATSRC and Fusion with LWT-UEO Optimization

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年12月15日

查看全文
  • Paluck Arora,
  • Rajesh Mehta,
  • Pramod Kumar Soni

摘要

Accurate detection of brain tumors in medical images remains a significant challenge, primarily due to the limited diagnostic capabilities of single-modality imaging, which often fails to capture the structural and textural complexity of brain tissues. To overcome these challenges, this paper introduces MultiRegFuse, a robust dual-step framework for multimodal medical image registration and fusion. The proposed method enhances both visual quality and diagnostic precision by integrating feature-based registration with fusion mechanism. In the registration phase, features are extracted using speeded-up robust features (SURF) and refined through alpha trimmed spatial relation correspondence (ATSRC), enabling precise alignment of reference and target images through homography transformation. Following registration, fusion stage employs lifting wavelet transform (LWT) to decompose the images into approximation and detail coefficients. A bidirectional two-stage fusion strategy is applied: F1-fusion merges approximation coefficient of reference image with detail coefficient of register image, while F2-fusion merges approximation coefficient of register image with detail coefficient of reference image, capturing complementary structural and textural information. To determine the optimal contribution from F1-fusion and F2-fusion, united equilibrium optimizer (UEO) algorithm is employed to compute the optimal fusion weights that enhance image quality. The fused coefficients generated through these weighted combinations are then passed through inverse LWT to reconstruct the final fused image. Experimental validation on both standard and clinical multimodal datasets demonstrates that MultiRegFuse consistently outperforms existing techniques across evaluation metrics such as SSIM increased from 0.7502 to 0.92873, PSNR increased from 18.8808 to 41.7691, Entropy increased from 7.423 to 12.4829, Correlation Coefficient improved from 0.9302 to 0.98621, and Mutual Information improved from 5.7087 to 9.45701 confirming its efficacy and reliability for clinical applications in brain tumor diagnosis.