Abstract
Purpose: :
Quality Assessment of digital fundus images is an important step in applications such as screening for ocular disease and multi-center clinical studies. For efficiency and consistency purposes there is a need for quantitative and objective grading methods for quality. We developed a computer algorithm based on human vision processes and pilot tested it on optic nervehead images.
Methods: :
‘Red-free’ fundus images (green) were utilized since they have been considered optimal in many cases. The study involved 59 eyes of 13 normal subjects and 35 glaucoma patients. The optic disc region was imaged without mydriasis with a camera designed for screening in the primary care environment (Digiscope GenIII). The images were graded without operator intervention by a dedicated algorithm. The latter was developed to mimic the procedure used by human observers, which was found to be based on evaluation of contrast and details in small areas of the image. The software program consisted of mathematical operations on image matrices performed on a MatLab (Mathworks ©) platform, concluding with the assessment of variance. The same set was graded by masked readers into four category (good, adequate, fair, poor). As a second task the algorithm was asked to identify the best image in a series of 10 consecutive frames acquired during a focusing procedure. A reader then identified the best image in the same series.
Results: :
The four subjective grades were compared to the numerical index derived by the algorithm. The sensitivity and the specificity of identifying poor images was 100 % and 100%, respectively and 92% and 78%, respectively, for fair images. All the best frames in a series of 10, identified by the reader, were within ±1 frame from that detected by the automated algorithm.
Conclusions: :
This pilot study demonstrates the feasibility of using a computer algorithm to grade the quality of digital fundus images. It also illustrates that the algorithm can be used successfully in automated focusing. Further studies are necessary to test if the algorithm can be applied to images obtained by other modalities
Keywords: image processing • imaging methods (CT, FA, ICG, MRI, OCT, RTA, SLO, ultrasound) • imaging/image analysis: non-clinical