Contributions of the inner retina and brain metabolites to glaucoma using information gain assessment

Abstract

Purpose : Recent studies have found the involvement of brain metabolites in both human and experimental animal models of glaucoma. However, it is unclear how important such metabolic changes are contributing to the disease. Here, we analyzed clinical ophthalmic assessments and brain magnetic resonance spectroscopy data for early glaucoma, advanced glaucoma, and healthy control subjects using information gain assessment.

Methods : Seventeen early glaucoma (age = 65.3 ± 8.2 years), 25 advanced glaucoma (age = 64.7 ± 10.4 years), and 22 healthy control subjects (age = 64.7 ± 8.0 years) (mean ± S.D.) underwent spectral-domain optical coherence tomography, Humphrey visual field perimetry, and proton magnetic resonance spectroscopy using a 3-Tesla magnetic resonance imaging scanner. To select the most useful descriptors for class selection, we ranked all descriptors in terms of their information gain using InfoGain with 10-fold cross-validations in the Weka software. We also ran the machine learning algorithm multi-layer perceptron (MLP) to determine the accuracy and area under receiver-operating-characteristics (ROC) curve using the reduced list of crucial attributes including peripapillary retinal-nerve-fiber-layer (pRNFL) thickness, macular ganglion cell-inner plexiform layer (mGCIPL) thickness, cup-to-disc ratio, visual field mean deviation, and choline, glutamate, and gamma-aminobutyric acid (GABA) in the visual cortex.

Results : When comparing healthy subjects to early or advanced glaucoma, pRNFL thickness ranked the first in terms of information gained. Interestingly, when comparing healthy subjects to early glaucoma, we found that choline metabolite in the visual cortex ranked the second, whereas GABA ranked the ninth for healthy subjects against advanced glaucoma. MLP was able to differentiate between: healthy control and all glaucoma subjects (accuracy: 85.9%; ROC: 0.913); healthy control and early glaucoma (accuracy: 76.9%; ROC: 0.778); and, healthy control and advanced glaucoma (accuracy: 91.5%; ROC: 0.969).

Conclusions : These results provide insight into the relationships between the eye, brain, and glaucoma, as well as the potential contributions of choline to early glaucoma and GABA to advanced glaucoma.

This abstract was presented at the 2023 ARVO Annual Meeting, held in New Orleans, LA, April 23-27, 2023.