No reference image quality assessment of retinal image for diabetic retinopathy detection based on feature extraction

Abstract

Retinal imaging is widely used to detect diseases like diabetic retinopathy, glaucoma and other diseases. One important factor affecting the detection outcome is the quality of the retinal image. Therefore, retinal image quality assessment is necessary as a preliminary detection. A method for evaluating the quality of the retinal image without reference (no reference) should be developed as the reference image is not always available. Additionally, a lot of research still using whole retinal image for assessment. In fact, the region of interest (RoI) for each disease to be detected is different from one another. This means that it is not the entire area of the retinal image to assess. This research developed a method for assessing the quality of the retinal image by cropping the RoI of the retinal image based on diseases to be detected, in this case it is focused on diabetic retinopathy. Retrieve the image features and determine the retinal image quality level by grouping them together using clustering techniques is presented in this paper. In the case of diabetic retinopathy, the best performance is to extract a combination of histogram characteristic, GLCM characteristic, and blood vessel contrast with a specificity of 77.36% and an accuracy of 72.41%.