Scholarlink

<p>Multimodal image registration plays a crucial role in biomedical research, enabling the integration of complementary information from different imaging techniques. We present a novel feature-based approach for multimodal image registration, alongside traditional intensity-based methods. Our method, inspired by SPP-net architecture, employs multi-level feature extraction for robust image alignment. Additionally, we perform t-SNE dimensionality reduction on the MALDI-MSI dataset to enhance feature discrimination and visualization. We evaluated both approaches using datasets from the ANHIR Grand Challenge and mass spectrometry imaging modalities (LA-ICP-MS and MALDI-MSI). The proposed feature-based method achieved comparable accuracy to optimized intensity-based approaches, with Dice Coefficients of 0.95 for ANHIR samples (e.g., COAD_05) and 0.97 for mass spectrometry data, while requiring approximately 50% less computational time. Quantitative evaluation through Mutual Information metrics and Hausdorff Distance demonstrated high registration accuracy across different tissue types and imaging modalities. These results establish our feature-based approach as an efficient alternative to traditional intensity-based methods for multimodal image registration in biomedical applications.</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

Novel feature-based method for multi-modal biomedical image registration compared to intensity-based technique

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年8月1日

查看全文
  • Mohammad Javad Shojaei,
  • Lichen Yang,
  • Kazem Shojaei,
  • Jeerapat Doungchawee,
  • Richard W. Vachet

摘要

Multimodal image registration plays a crucial role in biomedical research, enabling the integration of complementary information from different imaging techniques. We present a novel feature-based approach for multimodal image registration, alongside traditional intensity-based methods. Our method, inspired by SPP-net architecture, employs multi-level feature extraction for robust image alignment. Additionally, we perform t-SNE dimensionality reduction on the MALDI-MSI dataset to enhance feature discrimination and visualization. We evaluated both approaches using datasets from the ANHIR Grand Challenge and mass spectrometry imaging modalities (LA-ICP-MS and MALDI-MSI). The proposed feature-based method achieved comparable accuracy to optimized intensity-based approaches, with Dice Coefficients of 0.95 for ANHIR samples (e.g., COAD_05) and 0.97 for mass spectrometry data, while requiring approximately 50% less computational time. Quantitative evaluation through Mutual Information metrics and Hausdorff Distance demonstrated high registration accuracy across different tissue types and imaging modalities. These results establish our feature-based approach as an efficient alternative to traditional intensity-based methods for multimodal image registration in biomedical applications.