Scholarlink

<p>Gastrointestinal (GI) tract is a 9-meter-long food passage that transports food and nutrition to the human body. Due to the narrow space (~ 2&#xa0;cm in diameter) of the GI tract, a lesion removal surgery through endoscopic procedures faces limited view of the lesion, leading to the challenge of maximum removal of diseased lesion and minimum sacrifice of healthy tissues. This paper evaluates the application of the state-of-the-art AI technique of neural radiance field to reconstruct 3D models of a lesion based on 2D endoscopic videos. In this way, a full 3D view of concerned GI lesion can be presented to surgeons to allow comprehensive surgical planning. The advanced AI techniques, neural radiance field (NeRF), structure-from-motion and multi-view stereo are applied. This system is implemented based on COLMAP and Nerfstudio libraries. The system is evaluated using two sets of video clips containing 2600 images. Initial results illustrate that this end-to-end deep learning architecture, i.e. from 2D video input to 3D model output, presents considerable potential for reconstruction of GI lesions. The similarity measures of SSIM, PSNR and LPIPS between original (ground truth) and rendered images are of 19.46 ± 2.56, 0.70 ± 0.054, and 0.49 ± 0.05 respectively. This work contributes to the solution towards one of the long-standing challenges in computer vision field, which is to construct 3D views from 2D videos in the field of GI surgery planning and has achieved a promising performance. Future work includes enlarging datasets and removal of ghostly artefact from rendered images.</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

Neural Radiance Field Based 3D View Reconstruction for Gastrointestinal Tract Surgery Planning

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月15日

查看全文
  • Xiaohong W. Gao,
  • Annisa Ristya Rahmanti,
  • Barbara Braden

摘要

Gastrointestinal (GI) tract is a 9-meter-long food passage that transports food and nutrition to the human body. Due to the narrow space (~ 2 cm in diameter) of the GI tract, a lesion removal surgery through endoscopic procedures faces limited view of the lesion, leading to the challenge of maximum removal of diseased lesion and minimum sacrifice of healthy tissues. This paper evaluates the application of the state-of-the-art AI technique of neural radiance field to reconstruct 3D models of a lesion based on 2D endoscopic videos. In this way, a full 3D view of concerned GI lesion can be presented to surgeons to allow comprehensive surgical planning. The advanced AI techniques, neural radiance field (NeRF), structure-from-motion and multi-view stereo are applied. This system is implemented based on COLMAP and Nerfstudio libraries. The system is evaluated using two sets of video clips containing 2600 images. Initial results illustrate that this end-to-end deep learning architecture, i.e. from 2D video input to 3D model output, presents considerable potential for reconstruction of GI lesions. The similarity measures of SSIM, PSNR and LPIPS between original (ground truth) and rendered images are of 19.46 ± 2.56, 0.70 ± 0.054, and 0.49 ± 0.05 respectively. This work contributes to the solution towards one of the long-standing challenges in computer vision field, which is to construct 3D views from 2D videos in the field of GI surgery planning and has achieved a promising performance. Future work includes enlarging datasets and removal of ghostly artefact from rendered images.