Artificial Intelligence System Can Detect Glaucoma Severity

Ophthalmology. Visual Filed Testing. Automatic Computer Perimeter
Ophthalmology. Visual Filed Testing- Automatic Computer Perimeter.
Investigators say the system is “highly suitable” for both research and clinical practice.

A machine learning model can identify and delineate glaucoma presentations according to a 4-stage severity scale, reports an article published in the Journal of Glaucoma

Researchers selected participants from a Massachusetts Eye and Ear (MEE) dataset to create an objective severity staging system for glaucoma progression and used an independent data set to validate the results. To develop the staging system, researchers used k-means from the MEE dataset in an unsupervised clustering approach to identify clusters with similar visual fields (VFs) followed by post-hoc analyses to determine the statistical clusters using average mean deviation (MD) of each cluster. 

The researchers transformed statistical clusters into VF severity stages with a supervised analysis based on Bayes minimum error classification. They repeated this process using the independent dataset to validate the results. 

13,231 VFs from 8077 participants were taken from theMEE dataset and another 8024 VFs from 4445 participants from an independent data set for this study.

Four clusters of VFs were identified (6784, 4034, 1541, and 872) with average MDs of 0.0±1.4 db, -4.8 dB±1.9, -12.2±2.9 dB , and -23.0±3.8 dB, respectively, using the unsupervised clustering approach. 

Optimal MD thresholds to differentiate normal eyes from early, moderate, and advanced stage glaucoma were determined to be -2.2 dB, -8.0 dB, and -17.3 dB, respectively. This was determined during the supervised analysis. 

Researchers report a 94% accuracy for this staging system, describing it as “unbiased, objective, easy-to-use, and consistent,” and suggest it is “highly suitable for use in glaucoma research and for day-to-day clinical practice.”

Both the MEE and the independent dataset were collected in retrospective, cross-sectional studies, limiting researchers’ access to information regarding glaucoma severity. Additionally, the datasets were also missing information regarding possible comorbidities that could influence the results. Researchers also state that they did not create this severity staging scale with the intention of using it for diagnosis.


Huang X, Saki F, Wang M, et al. An objective and easy-to-use glaucoma functional severity staging system based on artificial intelligence. J Glaucoma. Published online June 3, 2022. doi:10.1097/IJG.0000000000002059.