Inverse Reinforcement Learning Intra-Operative Path Planning for Steerable Needle.

Area of Science:

• Neurosurgery
• Robotics
• Medical Simulation

Background:

• Keyhole neurosurgery demands advanced planning tools for flexible robotic systems.
• Intra-operative environments are dynamic, requiring real-time adaptation.
• Current methods may lack the precision and adaptability needed for complex procedures.

Purpose of the Study:

• To develop and evaluate a safe and effective intra-operative planning framework for flexible neurosurgical robots.
• To support neurosurgeons in adapting to dynamic environments during procedures.
• To enhance the precision and safety of steerable needle navigation in neurosurgery.

Main Methods:

• Integration of an inverse reinforcement learning path planning algorithm.
• Development of a pre-operative path planning framework for user interaction.
• Implementation of a realistic, time-bounded simulator using Position-based Dynamics (PBD) to model brain deformation.
• Simulation of a robotic system for testing the planning framework.

Main Results:

• The inverse reinforcement learning planner accurately guided a steerable needle to a target pose with low positional (1.34 ± 0.52 mm) and orientational (3.16 ± 1.06 degrees) error.
• The system demonstrated a 100% success rate in a deformable simulated environment.
• Re-planning time was rapid at 0.02 seconds, ensuring real-time adaptability.

Conclusions:

• The presented intra-operative steerable needle path planner effectively avoids anatomical obstacles while optimizing surgical criteria.
• The framework provides a fast, accurate, and robust solution for navigating flexible needles in complex neurosurgical procedures.
• This technology enhances safety and efficacy in keyhole neurosurgery by supporting real-time adaptation and precise instrument control.