Premium
Enhanced visualization of retinal vasculature using image processing
Author(s) -
Jung DoHyun,
Taek Kim Yun
Publication year - 2019
Publication title -
acta ophthalmologica
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.534
H-Index - 87
eISSN - 1755-3768
pISSN - 1755-375X
DOI - 10.1111/j.1755-3768.2019.5058
Subject(s) - retinal , artificial intelligence , diabetic retinopathy , mathematics , fluorescein angiography , kappa , fundus (uterus) , hue , computer vision , computer science , ophthalmology , pattern recognition (psychology) , medicine , diabetes mellitus , geometry , endocrinology
Objective To develop a digital filter that enhances visualization of retinal blood vessels. Methods Four‐hundred 24‐bit color fundus images were analyzed and properties of red, green, and blue channels were extracted. Then, using hemoglobin absorption coefficients, the relevant weights for gray‐scale conversion that emphasizes retinal vessels were calculated. To evaluate images, edges were detected via convolution‐al 2D Laplacian kernel from the processed images, and the number of edges, number of effective edges, and sum of intensities of edges were evaluated. To validate the clinical usefulness, two independent observers graded 100 fundus photographs from subjects with diabetes mellitus before and after image process described above. Agreement between observers and the diagnosis from fluorescein angiography was calculated using Cohen’s kappa coefficient (κ). Results The values of weights for red, green, and blue channels were calculated to be −0.0572, 0.7335, and 2.2079, respectively. When comparing the images that were processed using the new digital filter based on these values with the original image, gray‐scale, green, red, blue, and green and blue digital filter images, the number of edges, effective edges, and sum of intensities of edges were all found to be significantly higher in the images processed with the new filter (p < 10 −16 ). Cohen’s kappa coefficients (κ) of grading diabetic retinopathy from two independent observers comparing to the diagnosis from fluorescein angiography were 0.36, 0.29 respectively from fundus photographs and 0.61, 0.65 respectively from processed images. And Cohen’s kappa coefficients (κ) on the presence of any grade of diabetic retinopathy from two independent observers comparing to the diagnosis from fluorescein angiography were 0.69, 0.69 respectively from fundus photographs and 0.86, 0.84 respectively from processed images. Conclusions The RGB filter developed here was based on actual fundus images. The hemo‐globin absorbance reinforced the edges of retinal blood vessels, verifying that the new RGB filter can en‐hance the visualization of retinal vasculature. Significance: This digital filter was developed targeting hemoglobin for enhancement of retinal vasculature. Because this digital filter is simple to build, it can be applied for fundus camera.