Robust Path Planning via Learning From Demonstrations for Robotic Catheters in Deformable Environments.

Area of Science:

• Medical Robotics
• Computational Geometry
• Machine Learning

Background:

• Navigating tortuous and deformable vessels with catheters requires precise path planning.
• Current path planners often fail to adequately address the deformable nature of biological environments.

Purpose of the Study:

• To develop a robust path planning framework that accounts for catheter-vessel interactions and vessel deformability.
• To enhance catheter navigation accuracy in complex anatomical structures.

Main Methods:

• Proposed a novel path planner named Curriculum Generative Adversarial Imitation Learning (C-GAIL).
• C-GAIL utilizes a learning from demonstrations approach, considering the deformable properties of vessels and catheter-wall interactions.
• Conducted in-silico comparative experiments and in-vitro validation.

Main Results:

• C-GAIL achieved a 38% higher success rate in static and 17% higher in dynamic environments compared to GAIL.
• In-vitro validation showed a targeting error of 1.26 ± 0.55 mm and a tracking error of 5.18 ± 3.48 mm.
• These results represent 41% and 40% improvements over conventional centerline-following techniques.

Conclusions:

• The C-GAIL path planner demonstrates superior performance in deformable environments, outperforming state-of-the-art methods.
• The generated paths align better with catheter steering capabilities, enhancing user support for accurate navigation.
• The framework effectively manages vessel deformation uncertainty, leading to reduced tracking errors and meeting clinical accuracy requirements.