Depth-enabled fiber-bundle endoscopy for in situ intraluminal robotic navigation.

Area of Science:

• Medical Robotics
• Computer Vision
• Biomedical Imaging

Background:

• Three-dimensional (3D) perception is vital for intraluminal robotic navigation, requiring accurate depth estimation for safe advancement.
• Fiber-bundle endoscopy offers high-resolution, flexible imaging for miniature probes but struggles with reliable depth recovery due to artifacts and lack of ground truth data.

Purpose of the Study:

• To develop a weakly supervised framework to enhance fiber-bundle imaging with depth estimation capabilities.
• To enable accurate geometric feedback for robotic navigation within lumens without paired depth labels or external sensors.

Main Methods:

• A novel framework combining domain translation, pseudo-depth supervision, and structure-aware constraints was employed.
• The network was trained using only raw images, inferring depth directly and reconstructing lumen topology.
• The system was integrated with magnetic actuation for closed-loop control.

Main Results:

• The method successfully inferred depth from fiber-bundle images without ground truth data.
• Lumen topology reconstruction provided inclination-angle guidance for navigation.
• Efficient closed-loop navigation was demonstrated in curved luminal phantoms.

Conclusions:

• Depth-enabled fiber-bundle endoscopy can be achieved through weakly supervised learning.
• The proposed framework facilitates in situ robotic control for intraluminal procedures.
• This advancement holds potential for improved safety and efficacy in minimally invasive robotic surgery.