Currently, a glaucoma diagnosis only comes after as much as 40% of retinal ganglion cell (RGC) degeneration has already occurred and is observable through visual field changes—a sobering reality for newly diagnosed patients and their eye care providers.1 A new technology under investigation, DARC (detection of apoptosing retinal cells), aims to monitor this process of retinal cell death in vivo and catch the disease earlier than ever before.
DARC uses a fluorescently labeled protein that binds to a particular phospholipid in neurons to identify individual dying cells.2 An early study suggests the number of positively stained cells in a retinal fluorescent image can help researchers assess preclinical glaucoma activity and even glaucoma progression.2 However, this requires trained observers to manually count the spots on the DARC images.
A new study now shows artificial intelligence (AI) may be able to help. The researchers, who invented the DARC imaging technique, asked five observers to count the spots in the DARC images of 40 healthy controls and 20 glaucoma patients. Glaucoma eyes were followed for another 18 months and noted as stable or progressing based on patient eyes defined as progressing or stable based on OCT-retinal nerve fiber layer measurements at 18 months.
They then trained and validated a convolutional neural network (CNN)-aided algorithm using the manual counts from the controls and tested it on the glaucoma eyes.
They found the AI tool provided 97.0% accuracy, 91.1% sensitivity and 97.1% specificity when detecting spots compared with manual grading of the controls. For glaucoma eyes, the CNN-aided algorithm had 85.7% sensitivity. 91.7% specificity. The researchers added that both the observers’ and the algorithm’s DARC counts were significantly higher in those who were later found to be progressing at 18 months compared with those with stable disease. In addition, none of the stable glaucoma eyes had DARC counts above 30, suggesting this as a reliable threshold to separate stable eyes from those at risk of progression.
“This CNN-enabled algorithm provides an automated and objective measure of DARC, promoting its use as an AI-aided biomarker for predicting glaucoma progression and testing new drugs,” the researchers concluded in their study.
1. Kerrigan-Baumrind LA, Quigley HA, Pease ME, et al. Number of ganglion cells in glaucoma eyes compared with threshold visual field tests in the same persons. Invest Ophthalmol Vis Sci. 2000;41(3):741-8.
2. Normando EM, Yap TE, Maddison J, et al. A CNN-aided method to predict glaucoma progression using DARC (Detection of Apoptosing Retinal Cells). Expert Rev Mol Diagn. May 3, 2020. [Epub ahead of print].