Deep Video Analysis Methods for Surgical Skills Assessment in Cataract Surgery

It is important for a graduate surgical trainee in ophthalmology to have a strong understanding of how to proficiently perform cataract surgery. The surgical training curriculum should incorporate methodical assessments of surgical skills and improve trainee surgeons expertise to maintain patient safety. Prior rating scales for cataract surgery are highly dependant on the subjective opinion of the observing grader and are time consuming. This project is intended to develop a deep learning model for skill evaluation in cataract surgery using raw surgery videos that can supplement human review. An advanced convolutional neural network model is leveraged in this work and was evaluated using a large custom dataset. Videos from four phases in cataract surgery were used to quantify the model performance. Our model yielded an average accuracy of 82% for all four phases of cataract surgery.