Purchase this article with an account.
Chen Gong, John P Kelly, Laura Trutoiu, Brian Schowengerdt, Steven Brunton, Eric Seibel; Measurement of Retinal Vessel Width in Tele-ophthalmology for Mobile Health Monitoring. Invest. Ophthalmol. Vis. Sci. 2018;59(9):4624. doi: https://doi.org/.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Changes in arteriole and venule width, as well as their ratios, can be indicators of hypertension. We test the feasibility in tele-ophthalmology using smartphones to assess retinal health, with challenges of small field of view (FOV) via undilated pupil; and the potential broader use of head-mounted mixed reality (MR) equipment providing unique opportunities to monitor retinal health outside the clinic.
Retinal vessels in small FOV (source) images, captured in tele-ophthalmology, are compared to a large FOV baseline (reference) image obtained in clinic over time. First source images are registered to the reference image by the proposed joint dimension-reduction/phase correlation method (Fig.1), which has a high accuracy for small FOV and poor quality images. Knowing measurement points on retinal vessels in the reference image, vessel width of corresponding locations in the source image is measured in these steps: A small region covering the measurement point is cropped for blood vessel detection. A Gaussian matched filter is used for vessel segmentation. After binarizing, the detected vessel is thinned with a morphological operation to obtain the vessel's centerline location and direction. Finally, vessel width perpendicular to the centerline around the measurement point is obtained.In the experiment, six largest arteries and veins are selected near the optic disk of two normal retina reference images: one fundus image (45° FOV, EyeRounds.org) and one SLO image (55° FOV spectralis, Heidelberg Engineering). 342 small FOV (10°) source images are simulated from the reference image (45° FOV) with slight rotation and different added noise (Poisson, salt & pepper, multiplicative, and Gaussian white noise).
With different noise levels, the linear regression of the vessel width measurement error and ratio error between central retinal artery equivalent and vein equivalent are obtained. Over 20 independent measurements, the width error is within 1 pixel, which is around 10% of the chosen vessel’s width; The ratio error is within 0.1, which is around 14% of the true ratio.
Based on this preliminary result, it is possible to detect changes in vessel width or ratio over time that are larger than the regression error compared to baseline. Using this method, the onset of retinal disease may be detected with a mobile retinal imaging device outside of the clinic.
This is an abstract that was submitted for the 2018 ARVO Annual Meeting, held in Honolulu, Hawaii, April 29 - May 3, 2018.
This PDF is available to Subscribers Only