Dry Eye Prediction using Smartphones

Title: Dry Eye Prediction using Smartphones

PI: Dr. Anand Balasubramaniam, Dr. Pallavi Joshi, Dr. Kaushik Murali (Sankara Eye Hospital)

PI: Mohit Jain, Nipun Kwatra, Akshay Nambi (Microsoft Research)

 Short description

To develop an end-to-end system for dry eye prediction from input images and videos of the eye captured using smartphones.

 Background

Dry eye is one the most widespread eye diseases, affecting 8% of the global population. It is caused by decreased quantity/quality of tears. With the increasing smartphone penetration, developing world population sees dry eye disease as an emerging threat -- 275 million people in India are expected to be affected by 2030. Even rural India is likely to see 17 million new patients every year.

 There are several tests to detect dry eye, such as Schirmer’s test, Ocular Staining test, Tear Break Up Time (TBUT) test and LipiView test. Schirmer’s test measures the quantity of tear production, Ocular Staining test determines ocular surface health associated with insufficient tear flow and excessive dryness using rose bengal and/or lissamine green staining, and TBUT test measures the quality of tear film stability using fluorescein sodium strip. Johnson & Johnson LipiView device determines dryness in the eye, by recording 19.1 seconds video of each eye and using color interferometry to determine lipid layer thickness.

 Our Plan

In this project, we aim to develop an end-to-end system to estimate lipid layer thickness, NIBUT (non-invasive tear break up time), TMH (tear meniscus height) and blink dynamics from input images and videos of the eye (20-60 seconds long) captured using smartphones. As part of the data collection, for each participant, we will collect pairwise video data – using LipiView and using our smartphone setups – in different conditions, including after washing eye with cold water and after watching a 5-mins video with minimal blinks. Two different smartphone setups will be used: (a) two-phone setup for estimating lipid layer thickness and blink dynamics, and (b)mire-based setup: for estimating NIBUT and TMH (as discussed in the slide deck). We will also be collecting Schirmer’s test, Ocular Staining test and TBUT test data. The data after removing all the PII (personally identifiable information) will be used to evaluate our proposed approach of estimating lipid layer thickness compared to the gold standard LipiView device.

 For details, please refer the attached slide deck.

 Other details

• Sample size: Overall 200 dry eye patients and 200 healthy eye patient, over multiple phases.

• Inclusion/exclusion criteria: Only patients with dry eye condition and healthy eye can participate in this study. The patient must not have any other eye condition(s) except dry eye.

• Repeat Test: Yes, after watching a pre-selected YouTube video for 5 mins with minimal blinks, after washing eyes with cold water, after using Refresh Tears, and after one hour.