Upright to supine image registration and contour propagation for thoracic patients
arxiv(2024)
摘要
A renewed interest in upright therapy is currently driven by the availability
of upright positioning and imaging systems. Aside from reduced cost, upright
positioning possibly provides clinical advantages. The comparison between
upright and supine particle therapy treatments can be biased through multiple
variables, such as differences in the target contouring on the two CTs. We
present a method for upright and supine CT registration and structures
propagation, and the investigation of an AI-based contouring tool for upright
images. Six paired 4DCTs from Proton Therapy Collaboration Group registry were
available from the Northwestern Medicine Proton Centre. Deformable image
registration (DIR) is challenged by the different patient anatomy between
postures, causing artefacts in the warped images. To achieve high quality
contour propagation, we propose the construction of a region of interest
covering the ribcage volume to overcome this problem. As no target contour
ground truth was available, the registration quality analysis (QA) was
performed on lung structures, for which dice score coefficient (DSC) and
average Hausdorff distance (AHD) is reported. The TotalSegmentator tool,
trained on supine dataset, was applied on upright images, verified against lung
structures and used as additional comparison for contour propagation. The
TotalSegmentator QA results in a maximum AHD of 2mm and a minimum DSC of 0.94.
An average AHD of 1.5mm and 1.6mm, and an average DSC of 0.95 and 0.94 were
obtained comparing the propagated volumes to manually contoured and AI
structures, respectively. All AHD values are smaller than the CT slice
distances. The developed framework allows for target propagation between
upright and supine images, defining the first step to compare upright and
supine therapy of thoracic patients and enabling the application of image
fusion techniques in the upright therapy field.
更多查看译文
AI 理解论文
溯源树
样例
生成溯源树,研究论文发展脉络
Chat Paper
正在生成论文摘要