Purchase this article with an account.
Mark Christopher, Li Tang, John H Fingert, Todd E Scheetz, Michael David Abramoff; Automated discovery of optic nerve head structural features from image and genetic data. Invest. Ophthalmol. Vis. Sci. 2014;55(13):4744.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
To discover novel associations between optic nerve head (ONH) structure, genetic factors, and disease by applying computational methods to analyze stereo fundus images and genotyping data.
Stereo fundus images captured from genotyped participants (n=1057) of the Ocular Hypertension Study were used to measure the structure of the ONH. A stereo correspondence algorithm, optimized for fundus images, was applied to the images, generating a 3D map of the ONH region for each participant. Principal component analysis (PCA) was applied to the maps to extract structural features. The relationships between ONH structural features and allelic state at several glaucoma-associated loci were then modeled using a linear discriminant approach to maximize the predictive power. This resulted in genotype-based ONH structural features representing an estimate of the contribution of each locus to ONH structure. The resulting features were evaluated based on the strength of their association with genotype and their utility in early prediction of glaucoma.
The ONH structural features exhibiting strongest associations with genotype (p << 0.05) were identified for loci in the genes SIX1/SIX6, ATOH7, CDKN2B, TLR4, and ELOVL5. Incorporating these features into a model used for early prediction of glaucoma resulted in substantial increase in predictive power compared to a baseline model using only PCA-based ONH structural features.
The contribution of glaucoma-associated genes to ONH structure was evaluated by applying computational methods to a large dataset. By using a model that incorporated both imaging and genetic data, novel associations between phenotype and genotype were revealed. The identified ONH structural features were significantly associated with genotype and improved performance of glaucoma prediction models. Future work will examine additional genetic loci in order to further improve models used to predict and track glaucoma.
This PDF is available to Subscribers Only