January 19, 2016
3 min read

Automated system may be similar as manual grading in detecting glaucoma

An automated detection system using features of the optic nerve head may have comparable sensitivity as current classification systems.

You've successfully added to your alerts. You will receive an email when new content is published.

Click Here to Manage Email Alerts

We were unable to process your request. Please try again later. If you continue to have this issue please contact customerservice@slackinc.com.

An automated system for the detection of glaucoma may have a comparable performance for classification as compared with color fundus images viewed by trained ophthalmologists, according to a study.

Because undiagnosed glaucoma is a large contributor to preventable blindness, a number of known risk factors could provide guidelines to potentially target at-risk patients for early detection and treatment. Algorithms trained to detect topographic changes suggestive of glaucoma from color fundus photographs could be employed to screen huge populations for the presence of glaucoma.

Jayanthi Sivaswamy, PhD, S.R. Krishnadas, DO, DNB, and colleagues defined an objective decision-making system for the detection of glaucoma using fundus image-based analytics.

“The study suggests there is a possibility of screening for glaucoma by a computerized automated detection system without trained or skilled human personnel involved, provided fundus images of persons to be screened are available,” Sivaswamy told Ocular Surgery News. “This will allow for automated screening of populations for glaucoma on a large scale without human intervention, especially when skilled personnel for the purpose of glaucoma screening is scarce.”

Automated detection system

The automated system, described in the Journal of Glaucoma, is a computer-based application that can assess the cup-to-disc ratio, peripapillary atrophy, retinal nerve fiber defects and other topographic features of the optic nerve head in color fundus images. It provides a decision whether the optic disc in question is likely to be glaucomatous or normal.

“Persons who have been detected as possible glaucoma need to be evaluated by an ophthalmologist to exclude or confirm glaucoma,” Sivaswamy said.

A total of 1,926 eyes were used to train an automated system, 163 patients were clinically examined by two ophthalmologists, and 314 of their eye images were presented to four graders for their diagnostic decisions.

Clinical reference standard

The system had glaucoma assessment sensitivity and specificity of 0.716 and 0.717, respectively, with an area under receiver operating characteristic curve of 0.792.

Based on the four specialists, the average sensitivity and specificity at the eye level were 0.72 and 0.83, respectively, for the manual grading, and 0.72 and 0.72, respectively, for the automated system.

“The advantage of the automated classification system is that it can screen large populations without the need for trained or skilled human personnel,” Sivaswamy said. “If deployed widely, it can also become a very cost-effective screening solution for glaucoma.”

On the other hand, there is a possibility that early glaucoma may be missed by the automated system, and false negatives “may provide a false sense of security to persons screened, with the possibility of significant pathology being missed with opportunities lost for treatment or intervention,” she said.

Future research

Next, Sivaswamy would like to develop software tools to assess optic nerve heads based on disc photographs obtained by more innovative, less expensive cameras, similar to that of smartphone cameras or portable cameras, while also conducting a larger study for validity of the automated system.

“The system needs to be validated on a large sample of populations with varying ethnicity, population characteristics and optic disc morphology,” she said. “We also need to evolve and assess computerized automated detection that will take into consideration other risk factors in glaucoma, like the age, family history, level of intraocular pressure, etc., and integrate these risk factors with assessment of disc topography to be able to predict likelihood of glaucoma in the screened populations better.”

With further research and development, Sivaswamy hopes the automated system could someday replace routine and manual screening for glaucoma.

“It will also be increasingly possible to have a significant proportion of populations screened in non-hospital settings and enhance referral of those with suspected glaucoma,” she said. “With advances in information technology and telemedicine facilities, it will even be possible someday to be able to remotely screen or follow up patients with known glaucoma by ophthalmologists.” – by Kristie L. Kahl


Chakrabarty L, et al. J Glaucoma. 2015;doi:10.1097/IJG.0000000000000354.

For more information:

Jayanthi Sivaswamy, PhD, can be reached at the International Institute of Information Technology Hyderabad, Gachibowli, Hyderabad, Andhra Pradesh 500032, India; email: jsivaswamy@iiit.ac.in.

Disclosure: Sivaswamy reports no relevant financial disclosures.