Colin Goudie1,2, Keylor Chaves3, Ognjen Arandjelovic3, William Williams4, Andrew Blaikie3

Purpose: When attached to a mobile phone camera the Arclight Ophthalmoscope (AO) can acquire reti- nal video segments, offering an ultra-low cost opportunity to expand telemedicine in low-income settings. Innate limitations of direct ophthalmos- copy, i.e. narrow field of view, limit clinical utility, particularly in screening for diabetic retinopathy and retinopathy of prematurity. We aimed to create a novel computer algorithm capable of extracting images from videos providing wider-field fundal montages.
Methods: A novel algorithm to stitch video frames together was developed. Frames were analysed to detect the area of interest (retina) and discard arte- fact (iris, sclera & lids). Blood vessels provide land- marks to combine two frames, whilst removing corneal reflections by overlaying corresponding reti- nal content. This process is repeated for all suitable frames to create a single, wider-field still image, free of aberrations.
Results: 28 retinal videos through dilated pupils of healthy volunteers, acquired using an Arclight/ iPhone 6 combination, were analysed. Processing the 20-30 second video took 4-5 minutes using an Intel Core i5 processor with 8GB of RAM. Adequate montage was only possible where video continu- ously followed blood vessels (11/28 videos). The algorithm was robust and did not make assumptions regarding length, frame rate, size, or resolution.
Conclusion: The algorithm presented is capable of creating wide-field fundal montages from good quality videos where continuous blood vessel struc- ture is seen. Although in its infancy and not without limitations, this approach offers promise in increas- ing the clinical utility of mobile phone acquired reti- nal video for screening of eye diseases in low- income settings.