Scholarlink

<p>Accurately identifying brain tumours in preoperative and intraoperative medical images poses significant challenges for neurosurgeons, especially when relying solely on visual examination. Image registration aids in effectively addressing this challenge by improving the alignment and accuracy of medical images. However, several conventional image registration methods often struggle to determine optimum rigid transformation parameters and fail to identify the global optima or approximate to global optima, which significantly increases computational time. To counter this challenge, a two-tier approach is presented for multimodality register image, integrating hybrid optimization approach with convolution neural network. During the initial stage, source and template image alignment is performed by determining optimum transformation using hybrid optimization, with similarity metric such as mutual information as fitness function. Next, VGG-19 is applied to enhance the quality and reliability of source and hybrid optimized image. Subsequently, dynamic inlier selection optimizes the feature matching process, thereby improving the robustness of registration. Finally, thin plate spline interpolation is used to calculate the transformation parameters for accurate image alignment. The proposed approach is evaluated on monomodal and multimodal medical images sourced from standard and real-world datasets. The results show significant improvements in RMSE decreased from 14.346 to 8.487, SSIM increased from 0.9332 to 0.9698, PSNR increased from 36.079 to 39.812, and CC increased from 0.9619 to 0.989, outperforming the performance of existing methods. The experimental results confirm that proposed framework offers a robust, accurate, and computationally efficient solution for multimodal medical image registration.</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

Brain Tumour Image Registration Using Hybrid Evolutionary Optimization and Deep CNN

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月29日

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

摘要

Accurately identifying brain tumours in preoperative and intraoperative medical images poses significant challenges for neurosurgeons, especially when relying solely on visual examination. Image registration aids in effectively addressing this challenge by improving the alignment and accuracy of medical images. However, several conventional image registration methods often struggle to determine optimum rigid transformation parameters and fail to identify the global optima or approximate to global optima, which significantly increases computational time. To counter this challenge, a two-tier approach is presented for multimodality register image, integrating hybrid optimization approach with convolution neural network. During the initial stage, source and template image alignment is performed by determining optimum transformation using hybrid optimization, with similarity metric such as mutual information as fitness function. Next, VGG-19 is applied to enhance the quality and reliability of source and hybrid optimized image. Subsequently, dynamic inlier selection optimizes the feature matching process, thereby improving the robustness of registration. Finally, thin plate spline interpolation is used to calculate the transformation parameters for accurate image alignment. The proposed approach is evaluated on monomodal and multimodal medical images sourced from standard and real-world datasets. The results show significant improvements in RMSE decreased from 14.346 to 8.487, SSIM increased from 0.9332 to 0.9698, PSNR increased from 36.079 to 39.812, and CC increased from 0.9619 to 0.989, outperforming the performance of existing methods. The experimental results confirm that proposed framework offers a robust, accurate, and computationally efficient solution for multimodal medical image registration.