An unsupervised deep learning method for identifying glaucoma progression patterns using longitudinal ganglion cell complex (GCC) scans

Abstract

Purpose : In this study, we use unsupervised deep learning to identify different glaucoma progression patterns from macular GCC scans and investigate their relationship with the visual field (VF) and clinical parameters.

Methods : We used 28574 OCT scans from 5031 patients with at least one International Classification of Diseases (ICD) code for glaucoma (H40.xx). Longitudinal pairs of ganglion cell complex (GCC) scans corresponding to two separate visits for each patient were registered using Nifty registration method to have the same camera angle and magnification. Pixel-wise difference maps were calculated for pairs of scans and used to train an autoencoder. The number of clusters was defined with the AIC and BIC scores, elbow point, and silhouette score. Difference maps were then clustered based on the Gaussian Mixture Model (GMM). We then computed the correlation between the difference maps belonging predominantly to each pattern and corresponding VF total deviations (TDs), intraocular pressure (IOP), and cup-to-disc ratios.

Results : Our cohort consisted of 58% females, 11.31% Hispanic, 79.27% White, 13.68% Black, 6.68% Asian, and less than 1% other races. The distribution of the time difference between base visits and follow-up visits was 1.16-year mean and 0.88-year std. We identified 7 optimal OCT progression patterns which demonstrated good regional correlation with VF TDs (Figure 1). We additionally found regions in OCT difference maps with strong statistical correlation with IOP (Figure 2(A)), CDR change (Figure 2(B)), and VF MDs change (Figure 2(C)).

Conclusions : We demonstrate the utility of an unsupervised artificial intelligence approach for identifying OCT-based glaucomatous progression patterns. We additionally show regional correlation with VF TDs and define the regions associated with cup-to-disc ratio, intraocular pressure, and mean deviation change. The results of this study may allow for earlier identification of glaucoma progression using OCT scans.

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