A Deep-Learning-Based Guidewire Compliant Control Method for the Endovascular Surgery Robot.

Area of Science:

• Robotics
• Medical Imaging
• Artificial Intelligence

Background:

• Endovascular surgery is high-risk due to limited vision and guidewire control challenges.
• Current endovascular surgery robot (ESR) systems lack integrated force perception and vision.
• Guidewire manipulation is critical but difficult in minimally invasive procedures.

Purpose of the Study:

• To develop a deep learning-based guidewire-compliant control method (GCCM) for ESR.
• To enhance guidewire operation safety and efficiency by combining vision and force feedback.
• To mitigate risks associated with guidewire interaction during endovascular procedures.

Main Methods:

• A deep learning model (GCCM-net) was developed for real-time detection of guidewire tip-to-vascular wall collisions.
• A real-time operational risk classification method (GCCM-strategy) was proposed, using GCCM-net outputs for robotic damping and speed control.
• The GCCM-strategy was compared to traditional force sensors for managing force-position asynchrony in flexible guidewires.

Main Results:

• GCCM-net achieved a best accuracy of 94.86 ± 0.31% in identifying collisions within a vascular phantom.
• The GCCM-strategy effectively reduced robotic running speed via virtual resistance when surgical risks were detected.
• Experiments with operators demonstrated that GCCM-strategy reduced average operating force by 44.0% and operating time by 24.6%.

Conclusions:

• Deep learning-based integration of vision and force feedback positively impacts ESR efficiency.
• The proposed GCCM offers a novel approach to enhance safety and performance in robotic endovascular surgery.
• GCCM-strategy effectively addresses limitations of force feedback systems in complex guidewire manipulations.