Abstract
Purpose:
To develop and validate a software algorithm to automatically determine if a portion of the ocular surface is uncovered during a blink.
Methods:
The LipiView system captures video of the ocular surface and eyelids at 60 frames per second, and provides analysis of the tear film thickness and blink dynamics. A software algorithm automatically identifies blink segments using a pupil identification method. Nonblink segments are passed onto the lipid layer analysis engine, while blink segments are analyzed frame by frame to find regions of the ocular surface uncovered by the eyelids. The algorithm specifically searches for regions of high brightness and low chroma, which usually appear on the sclera or in the specular reflection of the LipiView light source on the cornea. Each pixel is tracked temporally to determine if it was covered during any frame in the blink segment. If any pixels remain uncovered throughout the blink, then the blink is identified as a partial blink. The partial blink detection algorithm was used to analyze 37 clinical videos containing a total of 203 blink segments. A human observer checked the same videos frame-by-frame to determine whether each blink fully covered the ocular surface.
Results:
Of the 203 blink segments analyzed, the software algorithm and human observer were in agreement for 201 (99.0%). The human observer identified 83 blink segments as full, of which the software correctly identified 82 (98.8%). The software and human observer agreed on 119 of the 120 partial blinks (99.2%).
Conclusions:
The partial blink detection algorithm in the LipiView software is capable of correctly identifying partial blinks with 99 percent accuracy. A full blink is necessary for the superior eyelid to contact the Meibomian gland orifices and spread lipid from the tear meniscus across the ocular surface. Partial blinks may result in decreased lipid layer thickness or nonuniform distribution of lipids.
Keywords: 549 image processing •
526 eyelid •
550 imaging/image analysis: clinical