Sequential Endoscopic-image 3D Reconstruction using Structured-light and Neural Signed Distance Field with...

Area of Science:

• Medical Imaging
• Computer Vision
• Computational Geometry

Background:

• Accurate 3D shape measurement of internal organs is vital for endoscopic diagnosis and surgery.
• Existing 3D endoscopic systems face challenges with noise and accuracy, particularly in one-shot scanning methods.
• Traditional feature-based Simultaneous Localization and Mapping (SLAM) struggles with the texture-less nature of internal organs.

Purpose of the Study:

• To develop an advanced 3D shape measurement technique for endoscopic procedures.
• To improve the accuracy and robustness of 3D reconstruction in challenging internal environments.
• To enable precise shape and appearance estimation for enhanced clinical applications.

Main Methods:

• A novel approach using neural signed distance fields (neural SDF) with a pattern reflection model.
• Structured light is modeled as high-frequency point light sources, eliminating the need for decoding.
• Differentiable rendering is employed for simultaneous estimation of shape, surface properties, and camera pose.
• Extension to endoscopic SLAM by sequentially optimizing camera pose for wide-area reconstruction.

Main Results:

• The proposed neural SDF method achieves accurate 3D shape reconstruction and surface property estimation.
• The technique demonstrates robustness in challenging endoscopic scenarios, outperforming previous methods.
• Validated effectiveness using rendered colon models, showing superior performance in shape reconstruction.

Conclusions:

• The neural SDF approach offers a significant advancement in endoscopic 3D shape measurement.
• This method has strong clinical relevance for cancer diagnosis, computer-assisted interventions, and generating training data.
• The technique provides a robust solution for wide-area shape reconstruction in internal organs.