Author Description

Estimating the progression of glaucoma can be a challenging task as the rate of disease progression varies among individuals. The influence of other factors such as measurement variability and the lack of standardization makes the task even more cumbersome. Using optical coherence tomography (OCT), anatomical changes in structures of the eye, such as the retinal nerve fiber layer (NFL) or the macula, can be measured to detect glaucoma before any functional damage is done. Using a patient's scans from prior measurements, a generative deep learning model using the conditional GAN architecture was used to predict glaucoma progression over time. To measure similarity between OCT scans that were generated using this method and ones that were imaged during live sessions, an index was calculated using structural similarity index measure (SSIM). The objective of this research was to improve on the methods of deriving this similarity index for OCT images and break away from the tradition of relying solely on mean squared error (MSE) and SSIM. By first isolating the region of interest in each of the image pairs then using multi-scale SSIM (MS-SSIM) to generate the metric value, we were able to increase the average SSIM value to 0.97. In a future work, we look to find more effective ways of isolating the region of interest in OCT scans as well as more effective ways of measuring similarity.