Abstract
Purpose :
The purpose of this paper is to determine if a convolutional neural network (CNN) can be used to predict the contribution of facial contour induced visual field defects in glaucoma patients. This can be utilized to differentiate visual field loss from glaucomatous damage from visual field defects due to facial contour.
Methods :
Glaucoma patients were consecutively enrolled and underwent a 60-4 visual field test. A single photograph was taken of each subject. A CNN was used to create a three-dimensional reconstruction based on the two-dimensional image. A map of the intersection between the visual axis and the face on the 3D reconstruction was used to predict the location and extent of visual field defects which corresponded to an individual’s facial contour. A python script was then utilized to create a visual field which demonstrated the visual field defects due to facial contour. The predicted 60-4 and the actual 60-4 visual field were superimposed to determine the amount of visual field attributed to glaucoma.
Results :
14 glaucoma patients were included in this study. 28 eyes were included in this study. Each eye completed a 60-4 visual field. The predicted 60-4 visual field related to a patient’s facial contour was superimposed on the patient’s actual 60-4 visual field. The duration of the 60-4 visual field tests ranged from 8 minutes 17 seconds to 11 minutes 55 seconds. In 27/28 eyes, the predicted visual field loss coincided with areas of visual field loss on their actual 60-4 visual field. The total thresholds, which were generated as a summation of each individual threshold point, ranged from 854-1534 in the right eye and 949-1604 in the left eye. The range of visual field affected by facial contour was 0-7% with an average of 3.14%.
Conclusions :
This study demonstrates the utility of a CNN assisted 60-4 visual field in monitoring glaucomatous progression. This tool can help differentiate visual field loss from glaucomatous damage from facial contour. In this study, facial contour commonly corresponded to nasal visual field defects. This methodology may be helpful in determining if a visual field defect is due to glaucoma or facial contour, assist in accurate identification of glaucoma progression, and reduce variability of visual fields by accounting for the role of facial contour in the peripheral field.
This abstract was presented at the 2022 ARVO Annual Meeting, held in Denver, CO, May 1-4, 2022, and virtually.