Abstract
Purpose :
Clinical education in vitreoretinal microsurgery is primarily based on verbal guidance and feedback from instructors to trainees. However, there are no established quantitative, objective, or standardized metrics to assess differences between trainees and instructors intraoperatively. This experimental study tested the hypothesis that instrument maneuvers are predictive factors in evaluating surgical skills in vitreoretinal surgery.
Methods :
Fifty vitreoretinal procedures consisting of core vitrectomy, membrane peeling, and endolaser photocoagulation were used in this study. Forty procedures were solely performed by fourteen fellows, while ten procedures were performed by one attending surgeon. The procedures were loaded into an instance segmentation neural network that tracked and recorded the location and path traveled by the vitrector, membrane peeling forceps, endolaser photocoagulation tooltip, and retinal landmarks such as the optic disc and fovea (Figure 1). The University of Illinois - Chicago Institutional Review Board judged the use of this dataset to be exempt due to its retrospective and unidentified nature.
Results :
Fellow surgeons spent more time executing the tasks (Table 1, P < 0.05 for membrane peeling and endolaser application). Fellows also tended to place the vitrector and endolaser probes closer to the optic disc throughout the procedure (Table 1, 20.34% & 28.75%, P < 0.05) compared to the attending surgeon. The same was observed with the membrane peeling forceps, although without statistical significance (8.54%, P < 0.2). A higher travel distance of the vitrector was also observed among fellows during the core vitrectomy (21.50%, P < 0.05), with increased idle time throughout all phases.
Conclusions :
There are potential discrepancies in instrument maneuvers between fellow and attending surgeons when we analyze a standardized set of surgical tasks. We propose a method based on a neural network to track the spatial location of instruments and tissues in the surgeons’ field of view, translating the extracted data into metrics that can expose these divergences. Potential applications include the creation of objective and standardized performance indicators to evaluate surgeons’ learning curves.
This abstract was presented at the 2023 ARVO Annual Meeting, held in New Orleans, LA, April 23-27, 2023.