Diabetic retinopathy classification using retinal fundus images with Squeeze VGG-16 Net

Diabetic retinopathy is a chief complication of diabetes mellitus, which is instigated by high blood sugar in the body. Diabetic retinopathy leads to various problems, like vision loss, decreased quality of life, retinal detachment, refractive errors, cataracts, genetic disorders, eye injuries, and increased healthcare costs. Regular monitoring and effective management are crucial to avoid vision loss. Therefore, an efficient Squeeze Visual Geometry Group (VGG)-16 Net is established to classify diabetic retinopathy. Primarily, a retinal fundus image is preprocessed by the median filter. Next, lesion segmentation is utilized for separating various lesions, like soft Exudates, hard Exudates, hemorrhages, and microaneurysms by the Relation Transformer Network (RTNet). Afterwards, blood vessel segmentation is performed by RTNet. Then, during feature extraction, features like Artery to Veins ratio, area, entropy, eccentricity, kurtosis, perimeter, Local Gabor XOR Pattern (LGXP), Median Binary Pattern (MBP), and Local Optimal Oriented Pattern (LOOP) are extracted. Lastly, diabetic retinopathy classification is established by the Squeeze VGG-16 Net developed by merging SqueezeNet and VGG-16 Net. Moreover, the proposed Squeeze VGG-16 Net achieved a True positive rate (TPR), True negative rate (TNR), accuracy, and F1-score of 92.187%, 93.178%, 92.099%, 91.879% correspondingly.