William Morgan, Anmar Abdul-Rahman, Dao-Yi Yu,Martin Hazelton, Brigid Betz-Stablein,Christopher Lind
Retinal venous pulsation detection is a sub-jective but very useful sign of elevated intracranialpressure, venous obstruction and glaucoma. To date nomethod can objectively measure and identify pulsatingregions automatically.
We developed a photo-plethysmographictechnique utilizing measurements of ?uctuatinghaemoglobin light absorption during pulsation. Highresolution video-recordings of the optic disk and retinaare taken with or without an ophthalmodyna-mometer. Pulsation amplitude was calculated fromregions of the retinal image video-frames and pulsa-tion threshold detection was compared to two observ-ers drawing an outline of the pulsating areas whileviewing video-clips.
216,515 image segments were analysed from26 eyes of 17 subjects. Using data from each eye, theaverage area under the receiver operator curve (AU-ROC) was 0.94. With all data analysed together theAU-ROC was 0.89. The ideal threshold amplitude fordetection of any pulsating segment being that withmaximal sensitivity and speci?city, was 5.5 (sd 2.1,99% con?dence interval 4.4 to 6.6) units compared to12 units before any would be missed. A multivariatemodel demonstrated that ideal threshold amplitudeincreased with increased variation in video-sequenceillumination (p = 0.0000), younger age (p = 0.0465),but not video-frame movement (p = 0.1297) or use ofophthalmodynamometer (p = 0.1952).
Conclusion: This technique demonstrates accurateidenti?cation of retinal vessel pulsating regions withno areas identi?ed by 2 observers being missed withthe automated technique. The amplitude values arederived objectively and may be a signi?cant advanceupon subjective ophthalmodynamometric thresholdtechniques.