Abstract
Purpose :
Reliable quantification of corneal sensitivity is desirable for the diagnosis and treatment of ocular disorders affecting comfort and eye health. So far, there has been no standardized method or equipment available to objectively measure the corneal sensation threshold. A novel liquid-jet aesthesiometer(LJA, Brien Holden Vision Inst.) was recently developed to address this issue, however, due to the subjective nature of the patient response, results from this test can vary. The goal of this project is to assess the feasibility of utilizing an automated blink detector to identify the blink reflex response to a stimulus delivered by the LJA.
Methods :
The slit-lamp(Nikon FS2) mounted LJA propels tiny saline droplets of variable size onto the cornea to stimulate the mechanoreceptors. A JAI GO-5000m USB camera recorded about 500 frames of 8-bit grayscale images(1280x1024) at 120 frames/s under infrared illumination, capturing the upper eyelid movements prior to and post-stimulus.
Using standard image analysis tools(NI Vision Dev. Module 2012 & LabWindows/CVI 2012, National Instr.), the position of the eyelid margin was detected in each frame and plotted over time. Actual reflex blinks in response to the stimulus were identified by filtering out normal eye/lid movements and looking for a particular lid movement pattern within a narrow window 75-200 ms after stimulation.
Results :
In a pilot study, 44 stimulations of various intensity were presented to 10 healthy participants’ right eyes and corresponding lid margin position traces extracted. Of these, 6 had no visible twitch in response to a range of stimuli. In 41 of the trial sets(92%), the blink/no blink event was correctly identified. In the remaining 3 sets(8%), the algorithm failed to identify the blink event, mainly because poor image contrast prevented detection of the lid margin. Processing time for image analysis and blink reflex detection was less than 0.5 s.
Conclusions :
The results demonstrate that automated blink detection is feasible to monitor response to LJA stimulation. Current indications demonstrate subthreshold stimuli do not invoke a blink reflex. It is anticipated that the 92% success rate can be further increased by refining the algorithms. Further clinical investigations are required to determine how automated blink detection correlates with subjective perception responses.
This abstract was presented at the 2019 ARVO Annual Meeting, held in Vancouver, Canada, April 28 - May 2, 2019.