Purchase this article with an account.
Anran RAN, Christopher L. Chen, Xi Wang, Vincent Mok, Clement C. Tham, Tien Y Wong, Carol Yim-lui Cheung; Association Between Alzheimer’s Disease and Artificial Intelligence-based Glaucomatous Optic Neuropathy Score. Invest. Ophthalmol. Vis. Sci. 2021;62(8):2121.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
The associations between Alzheimer’s disease (AD) and glaucoma are inconsistent in the current literature. We aim to assess the association between AD and glaucomatous optic neuropathy (GON) using artificial intelligence (AI) predicted continuous scores from optical coherence tomography (OCT) volumetric scans.
This study included 2,269 volumetric optic disc OCT scans measured with Cirrus HD-OCT (Carl Zeiss Meditec, Dublin, CA) from 440 eyes of 233 AD subjects and 1,158 OCT scans from 271 eyes of 136 cognitively normal subjects. GON score was calculated from each OCT scan using a validated AI deep-learning algorithm. GON score was a continuous variable ranged from 0 to 1, and a more significant score represented a higher probability of GON. In addition, average and quadratic retinal nerve fiber layer (RNFL) thicknesses were also extracted from the OCT device.
There was no significant difference in AI-based GON scores between AD and cognitively normal subjects (0.28 ± 0.31 vs. 0.30 ± 0.33, p = 0.20). However, AD subjects had significant thinner RNFL average (85.97 ± 17.65 μm vs. 90.35 ± 13.71 μm, p < 0.001), temporal (67.39 ± 17.02 μm vs. 74.31 ± 21.35 μm, p < 0.001), nasal (66.06 ± 15.28 μm vs. 67.68 ± 12.12 μm, p = 0.002), inferior (106.46 ± 29.58 μm vs. 112.54 ± 23.91 μm, p < 0.001) and superior (103.96 ± 25.79 μm vs. 106.86 ± 21.31 μm, p = 0.001) thicknesses. A simple linear regression test showed that the association between AD and AI-based GON scores was insignificant (p = 0.199). GON score is equal to 0.297-0.015x(diagnosis) when diagnosis is 1 (AD) or 0 (normal). Pearson’s correlation showed that GON scores had moderate and inverse correlation with RNFL average (R = -0.524, p < 0.001), superior (R = -0.495, p < 0.001), and inferior (R = -0.556, p < 0.001) thickness, as well as weak and inverse correlation with RNFL temporal (R = -0.210, p = 0.001) and nasal (R = -0.220, p < 0.001) thickness. AD was only weakly and inversely associated with RNFL average (R = -0.125, p < 0.001), temporal (R = -0.174, p < 0.001), and inferior (R = -0.103, p < 0.001) thicknesses.
Although AD subjects had thinner RNFL thickness than cognitively normal subjects, AD is not associated with AI-based GON score. Further research is still warranted to confirm the null association between AD and GON.
This is a 2021 ARVO Annual Meeting abstract.
An example of OCT volumetric scan predicted as GON by the AI algorithm.
This PDF is available to Subscribers Only