Disease Progression in Ophthalmic Images via Flicker Overlay

Abstract

Purpose : Glaucoma, a leading cause of irreversible blindness globally, requires longitudinal monitoring to track disease progression and treatment response. Current approaches for assessing and tracking disease progression with fundus photographs are often variable and subjective. By using a flicker chronoscopy tool (FCT), 2D images from separate timepoints are registered, or aligned, allowing users to “flicker” between pairs (or a series) of images, improving agreement between clinicians and enabling them to visualize CDR progression more often.

Methods : A total of 200 image pairs from 200 patients diagnosed with glaucoma were preprocessed to crop the optic nerve head using YOLOv8, followed by registration with our EyeLiner registration tool. Quality assurance was peformed on the registered images to acquire the 20 best aligned image pairs. These images were loaded into our FCT and three clincians were assigned 3 tasks. First was estimating cup-to-disc ratio (CDR) on 40 randomly assorted single images. Second was a comparison task for determining CDR progression on 20 image pairs juxtaposed side-by-side; categories were stable (ΔCDR of 0.0), mild progression (ΔCDR of 0.0-0.15), and significant progression (ΔCDR of 0.15-0.30). Lastly, the three raters determined if there was CDR progression on the flicker task (same categories as the comparison task), where the raters were able to rapidly alternate, or “flicker”, between the 20 registered image pairs.

Results : Agreement on single image CDR estimation with average rater CDR showed a correlation coefficient of 0.92, 0.93 and 0.94 for three raters, while the inter-rater kappa was 0.36, 0.33 and 0.39 on the comparison task and 0.51, 0.49 and 0.37 on the flicker task. This shows high inter-rater agreement estimating CDR when looking at independent images, as well as improvement in agreement when using flicker instead of side-by-side comparison. Additionally, we are able to demonstrate that all three raters visualized progression more often on the flicker task than the compare task.

Conclusions : The ability to flicker between photos allows clinicians to visualize CDR progression more often, in addition to improving the inter-rater agreement when detecting progression, versus juxtaposing them side by side. Although this was specific to glaucoma, we envision that our FCT will be used to visualize disease status and progression for many ocular diseases across various image modalities.

This abstract was presented at the 2024 ARVO Annual Meeting, held in Seattle, WA, May 5-9, 2024.