C3VDv2 - Colonoscopy 3D video dataset with enhanced realism

Area of Science:

• Medical Imaging
• Computer Vision
• Robotics

Background:

• Spatial computer vision holds promise for enhancing colonoscopy diagnostic accuracy.
• Current development of these techniques is limited by the scarcity of 3D colonoscopy datasets for training and validation.

Purpose of the Study:

• Introduce C3VDv2, an enhanced high-definition 3D colonoscopy video dataset.
• Facilitate quantitative evaluation of 3D colon reconstruction algorithms with improved realism.

Main Methods:

• Captured 192 video sequences (169,371 frames) from 60 silicone colon phantom segments.
• Provided ground truth data including depth, surface normals, optical flow, pose, and 3D models for 169 videos.
• Included simulated screening videos and videos with colon deformations for comprehensive assessment.

Main Results:

• C3VDv2 offers enhanced realism, simulating challenging scenarios like debris, mucous, blood, and rapid motion.
• The dataset provides extensive ground truth for robust algorithm development and evaluation.
• Includes data for both quantitative and qualitative assessments of 3D reconstruction.

Conclusions:

• C3VDv2 significantly advances the development and evaluation of 3D reconstruction algorithms for colonoscopy.
• The dataset's realism enables more robust and representative training and validation processes.
• Facilitates the advancement of spatial computer vision applications in endoscopic procedures.