Abstract
Purpose :
Eye-tracking is a useful tool to determine the feasibility of telerobotic surgery given that it can be used to evaluate impairments in the performance of surgeons when stereoscopic vision is altered. We analyzed data from a simulated, stereoscopic distortion experiment to uncover ocular motility adaptations in response to vision-distortions.
Methods :
Based on eye movement data collected in a simulated experiment (38 subjects were asked to manipulate a joystick to drop a ball into a receptacle that was randomly placed in a field before them 20 to 40 feet ahead, +/- 5 feet horizontally and 7 feet above), a principal component analysis (PCA) was performed to determine which eye metrics most explained the variance in task performance, as measured through error from target. Additionally, the trials were sorted into early, middle, and late intervals, and a time-binned statistical analysis was conducted to evaluate how the subjects’ performance varied throughout the course of the experiment. Beyond this, a 1D Multilayer Perceptron (MLP) was used to regress (predict) error and classify the time bin of a trial using saccade duration, saccade amplitude, and average saccade velocity as features.
Results :
The PCA analysis indicated that saccade amplitude and duration explained most of the variance in task performance, with average loadings of .688 and .634, respectively. A two tailed t-test of the time-binned analysis revealed evidence of improved performance in the middle trials but a decline during the late trials as measured by average error. The difference in task error between each of the time bins (early-middle and middle-late) were statistically significant with a p-value of less than 0.0001. The MLP model used to regress error and classify time bins achieved 0.71 accuracy on classifying early groups from middle and late groups and found that saccade amplitude was the most predictive feature. Saccade duration enhanced model performance towards imbalanced data and improved AUC to 0.65. Using all three features to regress the error, the mean absolute error (MAE) decreased to 0.04.
Conclusions :
The analyses showed evidence of various adaptations in response to vision-distortions. Subjects showed a learning effect coupled with a fatigue effect, improving performance in the middle trials but worsening performance in the late trials. The analyses also indicated that saccade amplitude is most insightful in understanding variation in ocular motility.
This abstract was presented at the 2024 ARVO Annual Meeting, held in Seattle, WA, May 5-9, 2024.