Optical Coherence Tomography-Guided Robotic Ophthalmic Microsurgery via Reinforcement Learning from Demonstration

Area of Science:

• Ophthalmology
• Robotics
• Artificial Intelligence

Background:

• Ophthalmic microsurgery presents significant challenges due to the fine motor skills and visual acuity required.
• Intraoperative optical coherence tomography (OCT) enhances visualization but does not overcome surgeon's physical limitations.

Purpose of the Study:

• To demonstrate the feasibility of using an industrial robot with learning from demonstration and reinforcement learning for OCT-guided ophthalmic surgery.
• To evaluate the performance of a reinforcement learning agent in performing OCT-guided corneal needle insertions.

Main Methods:

• An industrial robot was trained using learning from demonstration and reinforcement learning.
• The system performed OCT-guided corneal needle insertions in an ex vivo model of deep anterior lamellar keratoplasty (DALK) surgery.
• The reinforcement learning agent's performance was compared against surgical fellows in mock trials.

Main Results:

• The reinforcement learning agent successfully performed OCT-guided corneal needle insertions in ex vivo human corneas.
• The trained agent outperformed surgical fellows in accurately reaching a target needle insertion depth during mock surgery trials.

Conclusions:

• The combination of learning from demonstration and reinforcement learning is a viable approach for autonomous robotic ophthalmic surgery.
• This technology has the potential to overcome physical limitations in microsurgery and improve surgical outcomes.