Deep Learning Model Can Reliably Detect Presence of Dry Eye Disease

Physician Examining Patient's Eyes
The prospective study found the algorithm was 85% accurate in diagnosing DED, but not as successful in corneal/conjunctival staining or Schirmer’s test.

A new deep learning model to autonomously detect dry eye disease (DED) achieved reliable diagnosis when compared with standard dry eye clinical tests, according to findings published in Clinical Ophthalmology.

Researchers used anterior segment optical coherence tomography (AS-OCT) images (n=27,180) from 151 eyes of 91 patients, which were used to train and test the deep learning model. They were compared with the standard of diagnoses from masked cornea specialist ophthalmologists. The researchers performed clinical dry eye tests on patients in the DED group to compare the model results, including tear break-up time (TBUT), Schirmer testing, corneal staining, conjunctival staining, and Ocular Surface Disease Index (OSDI). 

The study showed an accuracy of 84.6%, sensitivity of 86.3%, and specificity of 82.3% of the deep learning model in diagnosing DED with a positive likelihood ratio of 4.89 and negative likelihood ratio of .17. In the DED group, mean DED probability score was 0.81+0.23, and in the healthy group, DED probability score was 0.20+0.27 (P <.01). The accuracy of the deep learning model in diagnosing DED was significantly better than that of corneal staining, conjunctival staining, and Schirmer’s test (P <.05). The difference between the deep learning diagnostic accuracy and that of the OSDI and TBUT was not statistically significant. 

The researchers explain that, combined with AS-OCT devices, deep learning algorithms like that used in the present study are convenient due to their ability to be used by a wide variety of healthcare professionals. Additionally, they note that this method has other features that set it apart from traditional DED detection techniques, such as “autonomy, objectivity, quick duration, and the test is non-invasive.”

Study limitations include its relatively small dataset, which prevented the age-matching of the 2 study groups. Additionally, the lack of a standard in dry eye research necessitated a novel definition of dry eye for the present study.


Chase C, Elsawy A, Eleiwa T, Ozcan E, Tolba M, Abou Shousha M. Comparison of autonomous as-oct deep learning algorithm and clinical dry eye tests in diagnosis of dry eye disease. Clin Ophthalmol. Published online October 21, 2021.doi:10.2147/OPTH.S321764