Real Time Prediction of Sclera Force with LSTM Neural Networks in Robot-Assisted Retinal Surgery

Area of Science:

• Ophthalmology
• Robotic Surgery
• Artificial Intelligence

Background:

• Retinal microsurgery demands high precision, with risks of sclera damage from excessive tool-sclera contact force.
• Existing robotic assistants improve accuracy but do not prevent surgeon misoperations causing dangerous forces.

Purpose of the Study:

• To develop a real-time predictive method for detecting and alerting surgeons to potentially damaging tool-sclera interactions.
• To enhance patient safety in retinal microsurgery by mitigating risks associated with excessive force application.

Main Methods:

• A Long Short-Term Memory (LSTM) recurrent neural network was employed to predict surgical tool behavior.
• User behavior data, including sclera force and insertion depth, was collected using a mock retinal surgery setup.
• The LSTM network was trained on time-series data to forecast future tool-sclera interactions.

Main Results:

• The LSTM network accurately predicted sclera force and insertion depth 100 milliseconds in advance (95.29% and 96.57% accuracy).
• Auditory feedback based on predicted forces alerted surgeons to potential dangers.
• The system reduced the occurrence of unsafe sclera forces from 40.19% to 15.43%.

Conclusions:

• The developed AI-powered predictive system significantly enhances safety in retinal microsurgery.
• Real-time force prediction and auditory feedback are effective in preventing excessive sclera force application.
• This technology holds promise for reducing surgical complications and improving patient outcomes in delicate eye surgeries.