Identifying glaucoma progression early is key to treating this chronic disease. Typically, progression is measured by assessing changes in visual field sensitivity, but optic disc and retinal nerve fiber layer (RNFL) changes are also common, and these cannot be assessed with perimetric testing.

OCT is well-suited to assess structural changes and objectively quantify neural loss—but these devices are expensive, not usually available in most clinics and require specially-trained technicians. Fundus imaging, on the other hand, is widely available and less expensive.

A research team decided to train a deep learning model on fundus images and corresponding OCT scans to create a model that would predict changes only observable on OCT, such as RNFL thickness and neuroretinal rim measurements, hopefully obviating the need for an OCT device and offering an easier avenue for detecting early glaucoma progression. 

The retrospective cohort study included a test sample of 33,466 pairs of color fundus photos and OCT scans from 1,147 eyes of 717 patients. The model was then tested on an independent sample of eyes.

The deep learning model obtained both objective and quantitative estimates of RNFL thickness from fundus images that correlated well with OCT measurements. The researchers believe their model could potentially be used to monitor for glaucotamous changes over time.