Real-time dense stereo reconstruction using convex optimisation with a cost-volume for image-guided robotic surgery

Area of Science:

• Medical Robotics
• Computer Vision
• Surgical Navigation

Background:

• Accurate 3D depth reconstruction is crucial for guiding robotic-assisted minimally invasive surgery.
• Existing stereo reconstruction methods struggle with textureless endoscopic scenes and occlusions, limiting their surgical application.
• Developing robust 3D models is essential for enhancing surgical precision and safety.

Purpose of the Study:

• To develop a dense stereo reconstruction algorithm for accurate 3D depth estimation in endoscopic surgical scenes.
• To address limitations of current methods in handling textureless tissues and occlusions.
• To create a smooth 3D model while preserving critical depth discontinuities.

Main Methods:

• A novel dense stereo reconstruction algorithm utilizing convex optimization and a cost-volume approach.
• Implementation designed for efficient and effective 3D model generation.
• Validation through quantitative analysis on simulated and real phantom data with ground truth, alongside qualitative assessment on real surgical images.

Main Results:

• The proposed algorithm successfully reconstructs smooth 3D models with preserved depth discontinuities.
• Quantitative evaluations demonstrate superior performance compared to state-of-the-art methods.
• Qualitative results from real surgical images show promising accuracy and robustness.

Conclusions:

• The developed stereo reconstruction algorithm effectively addresses key challenges in endoscopic imaging for surgery.
• The method offers improved accuracy and smoothness in 3D model reconstruction.
• The algorithm's parallelizability enables real-time performance on modern graphics hardware, facilitating practical surgical applications.