The study suggests a platform that can be integrated into surgical ophthalmic microscopes can provide real-time audiovisual feedback to the surgeon during phacoemulsification cataract surgery.
New research investigated the potential real-time image processing with artificial intelligence tools to extra data and guide phacoemulsification procedures to prevent complications leading to worse visual outcomes.
A team led by Yannek Leiderman, MD, PhD, Department of Ophthalmology and Visual Sciences, University of Illinois at Chicago evaluated the ability of a deep neural network (DNN) to track the pupil, identify the surgical phase, and activate specific computer vision tools to aid a surgeon during phacoemulsification cataract surgery.
Their findings determined an artificial intelligence-based surgical guidance platform can provide real-time feedback to the surgeon and showed the potential to enhance their experience during phacoemulsification cataract surgery.
Its cross-sectional study evaluated de-identified surgical videos of phacoemulsification cataract operation performed by faculty and trainee surgeons at the University of Illinois Hospital and Health Services between July 2020 - January 2021.
A total of 10 stereoscopic videos of phacoemulsification cataract surgery were captured using a stereoscopic surgical microscope. Then, 6 surgical procedures were selected at random and used for DNN training, with 2 eyes (n = 200 frames) for validation and 2 eyes (n = 23,640 frames) as a subset for the final evaluation.
For the comparative evaluation, a total of 101 phacoemulsification cataract procedures (n = 10,100 frames) from the Cataract-101 dataset were used, as well as to assess the generalizability of the platform. Frames were received from the video source in a region-based convolutional neural network and in real-time located the pupil.
Outcomes for the study were considered the area under the receiver operator characteristic curve (AUROC) and area under the precision-recall curve (AUPR) for surgical phase classification. They used Dice score (harmonic mean of the precision and recall sensitivity) for detection of the pupil boundary. The processing speed of the computer vision tools were calculated based on frames per second (FPS) achieved during network runtime.
Further, they administered a usability survey to volunteer cataract surgeons previously unfamiliar with the platform.
According to the results, the region-based convolutional neural network model achieved AUROC values of 0.996 (0.970 - 0.999) for capsulorhexis, 0.972 (0.935 - 0.999) for phacoemulsification, 0.997 (0.981 - 0.999) for cortex removal, and 0.880 (0.718 - 0.999) for idle phase recognition. A mean performance decrease in the AUROC of 6.8% was observed when applied to the external data set.
In addition, a comparison of the real size of the pupil with the pupil area detected by the algorithm yielded mean DIce scores of 90.2% in the local data set and 85.4% in the external data set.
The surgical guidance platform achieved a mean processing speed of 137 FPS during the execution of capsulorhexis guidance, 80 FPS during activation of the tools for lens material removal during phacoemulsification, and 92 FPS for the coretex removal set of tools.
A total of 11 phacoemulsification cataract surgeons performed the post-hoc evaluation of the surgical guidance platform, with 8 respondents (72%) mostly or extremely likely to use the guidance tool for complex cataract procedures. Additionally 5 (45%) thought it useful for non complex cataract procedures.
“The approach used in this study demonstrated the feasibility of integration between surgical microscopes and artificial intelligence–based platforms to provide surgical guidance in real time,” investigators wrote.
The study, “Evaluation of Artificial Intelligence–Based Intraoperative Guidance Tools for Phacoemulsification Cataract Surgery,” was published in JAMA Ophthalmology.