Purpose
To computationally identify quantitative features of optic nerve head (ONH) structure and apply those features to model the contribution of ONH structure to glaucoma.
Methods
ONH structural features were identified using baseline measurements and stereo fundus photos collected from the participants of the Ocular Hypertension Treatment Study. The collected measurements included demographic markers (age, sex, ethnicity), clinical measurements (visual acuity, intraocular pressure, cup-to-disc ratio, central corneal thickness, visual field measurements), and outcomes (conversion to POAG). ONH structure was inferred from the stereo photos using our validated stereo correspondence algorithm. The primary modes of variation of ONH structure were identified using principal component analysis. Contributions of baseline measurements and POAG to ONH structure were modeled using linear discrimant analysis. This resulted in a set of ONH structural features known as structural endophenotypes (STEPs). Association of the STEPs were evaluated against demographic, clinical, and outcome markers. The ability of the STEPs to predict incident POAG was also assessed.
Results
The computationally-identified STEPs showed significant associations with several clinical markers (age, ethnicity, cup-to-disc ratio, central corneal thickness, visual acuity). Incorporating the STEPs into models of incident POAG prediction led to an AUC of 0.722, a significant increase over baseline models using only demographic markers (AUC = 0.599).
Conclusions
A novel computational method for analyzing ONH structure was developed and evaluated. The resulting objective, quantitative measurements of ONH structure are significantly associated with widely-used clinical markers and useful in prediction of incident POAG. Future work will incorporate available longitudinal data to determine the stability of STEPs and attempt to identify time-dependent trends that aid in detecting disease onset prior to any loss of vision.