Abstract
Purpose :
In situations where intraocular lens fixation presents challenges due to the lens capsule's support, intrascleral fixation is increasingly becoming a crucial technique. Despite its growing importance, there is a lack of established methodologies for the quantitative analysis and assessment of these surgical procedures. In our research, we sought to provide an objective evaluation of the intrascleral fixation technique. This was achieved through the utilization of a stretchable strain sensor in conjunction with an anterior segment model eye, enabling the analysis of hand movement waveforms during the procedure.
Methods :
We equipped all ten fingers with the C-STRETCH stretchable strain sensor (manufactured by Bando Chemical) and conducted the intrascleral fixation technique via the flange method on a model of the anterior eye segment. Throughout this process, hand movement waveforms were recorded 96 times from five medical professionals at our hospital, including both doctors and orthoptists, comprising three experts and two novices. We analyzed these waveforms to identify specific steps: the insertion of the intraocular lens' leading loop into the 30G needle (termed as Step 1), and the placement of the rear loop (Step 2). Normalization was applied to the length of each step's waveform, followed by dimensionality reduction through feature extraction. Using LightGBM, we classified these five surgeons and also differentiated between the experienced and beginner surgeons. The primary metric for evaluation was the overall accuracy achieved in a 3-fold Cross-Validation (K=3) setup.
Results :
In the classification of surgeons, the accuracy for Step 1 reached 71.9% (69 out of 96 cases), while in the binary classification, it achieved 86.5% (83 out of 96 cases). For Step 2, the accuracy rates were 69.8% (67/96) in surgeon classification and 75% (72/96) in the binary classification.
Conclusions :
Applying machine learning to analyze movement patterns revealed distinct characteristics for each surgeon involved in the study. This research indicates the feasibility of using model eyes and hand sensors for the objective assessment of surgical skills.
This abstract was presented at the 2024 ARVO Annual Meeting, held in Seattle, WA, May 5-9, 2024.