A CNN-based prototype method of unstructured surgical state perception and navigation for an endovascular surgery...

Area of Science:

• Robotics
• Artificial Intelligence
• Medical Technology

Background:

• Robot-assisted endovascular surgery (ES) performance is limited by surgeon skill dependency.
• Master-slave robotic structures present challenges for autonomous ES due to skill modeling and state perception difficulties.

Purpose of the Study:

• To develop a novel CNN-based framework for autonomous ES robot navigation.
• To address challenges in surgeon skill modeling and surgical state perception for autonomous ES.

Main Methods:

• A 2D CNN-based operating action probability estimator was developed for surgical state image feature extraction and action mapping.
• A 1D CNN with multi-input was used to recognize guide wire operating force conditions.
• An eye-hand collaborative servoing algorithm integrated network outputs for closed-loop robot control.

Main Results:

• The CNN-based framework demonstrated adaptability to different situations in real-world ES robot testing.
• The method achieved success rates and average operating times comparable to the state of the art.
• Robotic operations mimicked manual operation in trajectory and operating force levels.

Conclusions:

• The proposed CNN-based method effectively learns surgeon skills for autonomous ES robot control.
• This framework shows potential for easy extension to other surgical robots.
• The approach enhances autonomous capabilities in robot-assisted surgery.