Cost-Efficient Video Synthesis and Evaluation for Development of Virtual 3D Endoscopy

Area of Science:

• Medical Imaging
• Computer Vision
• Computational Geometry

Background:

• Accurate 3D reconstruction of hollow organs from endoscopic video is crucial for early cancer detection and surveillance.
• Evaluating 3D reconstruction algorithms requires extensive datasets, realistic simulations, and robust performance metrics.

Purpose of the Study:

• To develop a cost-effective simulation platform for synthesizing monocular endoscope videos and ground truths.
• To introduce a novel evaluation procedure for assessing and comparing 3D reconstruction pipelines.

Main Methods:

• A computer-aided simulation platform was developed to generate synthetic monocular endoscope videos and corresponding ground truths.
• A bladder phantom and cystoscopy were used as a model case to create an extensive dataset.
• A novel evaluation procedure was introduced to assess 3D reconstruction pipeline performance.

Main Results:

• The platform was used to evaluate a proprietary (CYSTO3D) and a general-purpose (COLMAP) 3D reconstruction pipeline for bladder cystoscopy videos.
• Evaluation results provided insights into clinical acquisition protocols and areas for pipeline refinement.
• The proposed toolset demonstrated efficient performance assessment and problem identification for 3D reconstruction pipelines.

Conclusions:

• The developed toolset enables efficient assessment of 3D reconstruction pipeline performance.
• It facilitates comparison between different pipelines and provides actionable insights for technical and clinical improvement.
• This work contributes a valuable resource for advancing 3D reconstruction in endoscopic applications for cancer diagnosis.