Abstract: Diabetic retinopathy is a consequence of Diabetes Mellitus that affects the Retina (back of the eye) due to excessive blood sugar levels. If left undiagnosed and untreated, it might result in blindness. The retina is the light-sensitive layer of cells that turns light into electrical signals at the back of the eye. The signals are transmitted to the brain, which transforms them into the images you see. The retina requires a constant supply of blood, which is delivered via a network of small blood capillaries. Advanced cases of diabetic retinopathy may necessitate a surgical treatment to remove and replace the vitreous, a gel-like fluid in the back of the eye. A retinal detachment may also necessitate surgery. This is a separation of the rear of the eye's light-receiving lining. Diabetic retinopathy (DR) diagnosis by colour fundus images necessitates skilled doctors recognizing the presence and significance of several minor characteristics, which, combined with a complicated grading system, makes this a challenging and time-consuming task. In this study, we present a CNN approach for diagnosing DR and reliably grading its severity from digital fundus images. [1-4] We create a network with CNN architecture and data augmentations that can recognize the complex elements involved in the classification task, such as micro-aneurysms, exudates, and haemorrhages on the retina, and deliver diagnosis automatically and without the need for human input. We train our network on the acquired images, which are applied with Gaussian filters. On 3500 validation photos, our suggested CNN achieves a sensitivity of more than 95% and an accuracy of 98% on the image set. The proposed method is extremely accurate at objectively diagnosing and grading diabetic retinopathy, removing the requirement for a retina specialist and increasing access to retinal care. This technique allows for early detection and objective tracking of disease progression, which may aid in the optimization of medical therapy to reduce visual loss. [5]
Keywords: Diabetes Mellitus, Deep Learning, Convolutional Neural Networks (CNN), Diabetic Retinopathy, Image Classification, retina, Gaussian filters, Mild DR, Moderate DR, Severe DR, Proliferate DR and NO DR.
|
DOI:
10.17148/IJARCCE.2022.115109
