Abstract
Purpose :
To find approaches to the accuracy in retinopathy of prematurity (ROP) diagnosis.
Methods :
Group #1 I stage (n=7, 7 eyes), group #2 II stage (n=6, 6 eyes), group #3 III stage (n=4, 4 eyes), group #4 IVA stage (n=3, 3 eyes), group #5 posterior aggressive ROP (PAROP) (n=2, 2 eyes), group #6 (control) - immature retina (n=6, 6 eyes).We use the best pix from RetCam Shuttle (Clarity MS, USA) video, modeling wide-field image, identifying missed the "mute" zones, the localization of the macula and checking the index of traction (Tm), the zone and extension, fractal dimension (Df) and complexity of vascular systems (CVS) (A-1 creation of a preliminary capillary plexus, B-2 normal vascularization, C-3 pathological vasculogenesis). Tm is width-to-length attitude of ellipse (optic nerve head to temporal branches of the retinal vessels). Our original method is a way of stitching images when examining children without anesthesia. Automatic wide-field fundus image algorithm: adaptive filtering, separation of the green channel, applying a Gaussian blur filter, local histogram equalization, the general transformation map. Support of the program for stitching images was at the first stage, the work was done due to the personal contribution of the researcher's. Statistical methods: Mann-Whitney U-test, Spearman's rank correlation coefficient (SPSS)
Results :
Result format: Median(25%-quartile;75%-quartile). Group #1 Df 1,31(1,3;1,34), Tm 0,79(0,79;0,83), CVS 1(1;1,25). Group #2 Df 1,38(1,36;1,39), Tm 0,76(0,71;0,78), CVS 1,5(1,5;2). Group #3 Df 1,46(1,45;1,47), Tm 0,68(0,64;0,78), CVS 2,5(2,5;2,63). Group #4 Df 1,55(1,55;1,56), Tm 0,62(0,62;0,66), CVS 3(3;3). Group #5 Df 1,66(1,65;1,66), Tm 1(1;1), CVS 3(3;3). Group #6 Df 1,27(1,27;1,27), Tm 0,95(0,91;0,99), CVS 1(1;1). Strong correlation between Df and stages (ρ=0,85, p=0,01); negative correlation Tm and stage (ρ=-0,62, p=0,01) except PAROP; strong positive correlation CVS and stage (ρ=0,91, p=0,001) were determined. Platform modules for a wide-field image for macula area, morphometry and isolation of the vascular nodules with neural network have been developed.
Conclusions :
The developed algorithm fits new ROP screening and treatment control criteria and can be used to identify the stage of the ROP as a holistic picture of the disease on one platform.
This is a 2021 ARVO Annual Meeting abstract.