Abstract
Purpose :
Dry eye disease (DED) is a multifactorial, chronic, and progressive disease that alters the ocular surface and the tear film. Millions of people in the world are affected by DED.Patients with DED have a deteriorated quality of life and visual function.One of the most common subtypes of DED is evaporative dry eye, directly related to the Meibomian Glands (MG).There are different objective and subjective tests that allow its diagnosis and classification.However, there is no Gold standard test or a definitive consensus among professionals about which is the ideal set of test for its diagnosis.In recent years, the use of artificial intelligence has represented a great advance in the field of biomedical and health sciences, allowing the development of promising techniques for aiding the prediction of diseases, from either numerical data or images.
The main objective of this study was the prediction of the diagnosis of dry eye disease, from objective and subjective indicators, using machine learning models.
Methods :
Relevant clinical tests to the diagnosis of dry eye have been carried out on 73 subjects (30 control,19 CL wearers and 24 with MG pathology). Symptomatology tests, ocular surface and MG recognition were performed. Using the data obtained in these tests, 24 machine learning classifiers were trained and the top 5 were verified.
Results :
Accuracies between 87-96% were obtained in 5 of the classifiers trained to differentiate the different study groups according to the tests performed. The precision obtained for the verification set decreased for all classifiers, obtaining the highest precision with 87.3 %reliability.
Conclusions :
Machine learning classifiers can be trained to classify a group of patients using clinical data. A significant reduction in accuracy has been observed when going from the training group to the verification group.An increase in the number of study subjects would mean an increase in the precision of the classifiers. Machine learning techniques can be useful aids for diagnosing multifactorial pathologies such as meibomian gland dysfunction or dry eye syndrome.
This abstract was presented at the 2023 ARVO Annual Meeting, held in New Orleans, LA, April 23-27, 2023.