Abstract
Purpose :
Turbulent intraocular fluid flow and brusque or rapid tool movement during phacoemulsification cataract surgery can lead to potential complications such as the rupture of the lens capsular bag. Quantifying these variables and providing feedback to the surgeon in real time, or the automated control of fluidic parameters using these data, may help to enhance the surgical environment.
Methods :
Segmentation of the pupil area was performed in real time by a neural network, then optical flow was used to track features inside the capsular bag. Under high-flow conditions during aspiration of lens fragments, the algorithm was able to calculate the movement of lens material and motion of tools located within the pupil area. The sensitivity threshold for motion estimation during the procedure may be varied according to the user preference. We evaluated the performance of our algorithm by comparing computed values with manual annotation of surgical videos by ophthalmic surgeons.
Results :
The experts’ annotation comparing the real size of the pupil with the pupil area detected by the algorithm resulted in an overall precision, recall and interception over union (IoU) of 82.07%, 87.19% and 95.14%, respectively.
Post-processed video streaming was achieved at rates greater than 60 frames per second, demonstrating feasibility for implementation in real-time. All ophthalmic surgeons invited evaluated the tools as useful, while 86% of the participants considered the features presented intuitive.
Conclusions :
Feedback to the surgeon relating to turbulent flow of lens fragments and/or brusque movements of microsurgical tools may potentially enhance the surgeon’s experience during cataract surgery. Additional studies are required to assess the feasibility for implementation into our current surgical paradigm.
This is a 2021 Imaging in the Eye Conference abstract.