Scholarlink

This study aims to creatively construct lung 4D-PET images by integrating precise 3D motion data from 4D-CT with rich functional information from 3D-PET. By applying advanced registration algorithms based on finite element method, we successfully extracted displacement vector field (DVF) from 4D-CT and performed fine reconstruction in the three-dimensional spatiotemporal dimension. Subsequently, this reconstructed DVF was applied to static 3D PET images, resulting in high-precision synthesized 4D-PET images. In order to verify the accuracy and reliability of synthesized 4D-PET, we conducted in-depth comparative analysis of tumor motion trajectories obtained from collected 4D-PET and synthesized 4D-PET in three dimensions: up and down (SI), front and back (AP), and inside and outside (ML). The results showed that the motion trajectories between the two were highly consistent, with correlation coefficients (CC) significantly exceeding 0.94 in all directions, and the mean difference in motion amplitude (Ds) remained below 0.74 mm, fully verifying the accuracy of the synthesized 4D-PET. In addition, we also used normalized cross-correlation (NCC) as an evaluation metric to quantitatively compare the image quality of synthesized 4D-PET with the collected 4D-PET. In the final inspiratory (EI) stage, the average NCC value of the synthesized 4D-PET was as high as 0.94 ± 0.02, indicating that its image quality reached a very high level, comparable to the original collected data. Furthermore, we accurately segmented the tumor and lung tissue volumes in the synthesized 4D-PET and compared and analyzed the segmentation results with the collected 4D-PET. The results showed good consistency in volume segmentation between the two, especially during the EI and median inspiratory phases. The Dice similarity coefficients (DSC) of the left lung, right lung, and tumor remained above 0.90, with the highest reaching 0.96 ± 0.01, fully demonstrating the reliability of the synthesized 4D-PET in volume quantification. In summary, this study preliminarily verified the feasibility and effectiveness of using deformable image registration (DIR) method to generate synthetic 4D-PET, providing a new technological path for accurate assessment and treatment monitoring of lung diseases.

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

Retrospective 4D-PET Synthesized Using Deformable Image Registration (DIR) Method

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Hui Peng,
  • Juan Yang,
  • Yanchao Lou

摘要

This study aims to creatively construct lung 4D-PET images by integrating precise 3D motion data from 4D-CT with rich functional information from 3D-PET. By applying advanced registration algorithms based on finite element method, we successfully extracted displacement vector field (DVF) from 4D-CT and performed fine reconstruction in the three-dimensional spatiotemporal dimension. Subsequently, this reconstructed DVF was applied to static 3D PET images, resulting in high-precision synthesized 4D-PET images. In order to verify the accuracy and reliability of synthesized 4D-PET, we conducted in-depth comparative analysis of tumor motion trajectories obtained from collected 4D-PET and synthesized 4D-PET in three dimensions: up and down (SI), front and back (AP), and inside and outside (ML). The results showed that the motion trajectories between the two were highly consistent, with correlation coefficients (CC) significantly exceeding 0.94 in all directions, and the mean difference in motion amplitude (Ds) remained below 0.74 mm, fully verifying the accuracy of the synthesized 4D-PET. In addition, we also used normalized cross-correlation (NCC) as an evaluation metric to quantitatively compare the image quality of synthesized 4D-PET with the collected 4D-PET. In the final inspiratory (EI) stage, the average NCC value of the synthesized 4D-PET was as high as 0.94 ± 0.02, indicating that its image quality reached a very high level, comparable to the original collected data. Furthermore, we accurately segmented the tumor and lung tissue volumes in the synthesized 4D-PET and compared and analyzed the segmentation results with the collected 4D-PET. The results showed good consistency in volume segmentation between the two, especially during the EI and median inspiratory phases. The Dice similarity coefficients (DSC) of the left lung, right lung, and tumor remained above 0.90, with the highest reaching 0.96 ± 0.01, fully demonstrating the reliability of the synthesized 4D-PET in volume quantification. In summary, this study preliminarily verified the feasibility and effectiveness of using deformable image registration (DIR) method to generate synthetic 4D-PET, providing a new technological path for accurate assessment and treatment monitoring of lung diseases.