Image Segmentation Using DeepLab V3+ for Diabetic Retinopathy

Numerous eye conditions affect diabetics, including diabetic retinopathy, glaucoma, and hypertension. These are the most prevalent eye conditions that pose a hazard to vision because of imposed changes to the vasculature. According to the technique suggested, neural network training may be used to accurately segment the retinal arteries. In this work, preprocessing methods such as Haar transform, and Discrete Cosine Transform have been proposed. The Haar transform compresses the retinal images to retrieve just the key components. The pre-processed images are then segmented using the two different CNN models to compare the accuracy. The segmented retinal images from the preprocessing have been created using U Net, DeepLabV3+ and Iterative UNet. Our experimental results, reveal intriguing insights into the model's comparative efficacy. Notably, Deeplab V3+ stands out as the top-performing model, achieving an outstanding accuracy of 96.767% when coupled with DCT preprocessing.