Purchase this article with an account.
Reiko Arita, Katsumi Yabusaki, Takanori Yamauchi, Tadashi Ichihashi, Naoyuki Morishige; Diagnosis of dry eye subtype by artificial intelligence software based on the interferometric fringe pattern of the tear film obtained with the Kowa DR-1α instrument. Invest. Ophthalmol. Vis. Sci. 2018;59(9):1965.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
We previously showed that the DR-1α tear interferometer (Kowa) has the potential to identify subtypes of dry eye on the basis of classification of interferometric fringe patterns as pearl-like appearance (normal tear condition), Jupiter-like appearance (aqueous-deficient dry eye), or crystal-like appearance (evaporative dry eye). We here set out to develop and validate automatic diagnosis software for analysis of tear interferometric fringe patterns obtained with DR-1α.
We constructed an artificial intelligence–based diagnosis (AI-diagnosis) model using 11 predetermined image features that distinguish the three types of interference fringe pattern obtained with DR-1α. The model was then trained with 138 images of each type, with each type being determined on the basis of the predominant pattern observed in fives images collected at 0.5-s intervals from 3.0 to 5.0 s after a blink. One hundred interferometric movies (33 pearl-like appearance, 33 Jupiter-like appearance, 34 crystal-like appearance) obtained from 100 control individuals or dry eye patients (54 men, 46 women; mean age ±SD, 42.3 ± 15.2 years) were analyzed for diagnosis of dry eye subtype on the basis of the interferometric fringe pattern either by a dry eye expert (R.A., IF-diagnosis) or by AI-diagnosis. The definitive clinical diagnosis was based on tear film parameters. The agreement between the clinical diagnosis, AI-diagnosis, and IF-diagnosis was assessed, and the F-score was calculated as a measure of the accuracy, considering both precision and recall, of AI-diagnosis.
We obtained kappa values of 0.760 and 0.714 for comparison of AI-diagnosis with IF-diagnosis or with clinical diagnosis, respectively. AI-diagnosis was found to be reliable, with F-scores of 0.954, 0.806, and 0.762 for Jupiter-like, crystal-like, and pearl-like pattern types, respectively.
Our AI-diagnosis software successfully identified dry eye subtype based on interferometric fringe patterns obtained with DR-1α. More precise diagnosis might be obtained on the basis of the combination of the noninvasive tear film breakup time determined with DR1-α and the interferometric fringe pattern.
This is an abstract that was submitted for the 2018 ARVO Annual Meeting, held in Honolulu, Hawaii, April 29 - May 3, 2018.
This PDF is available to Subscribers Only