Abstract: Diabetic retinopathy, which results from persistently high blood sugar caused by Diabetes Mellitus or simply called Diabetes, is linked to harm the microscopic blood vessels in the retina. In order to send signals to the brain via the optic nerve, the retina must first detect light. Vision distortion can result from diabetic retinopathy, which can cause blood vessels in the retina to leak fluid or haemorrhage (bleed). In its most severe form, aberrant blood vessels proliferate (grow in quantity) on the retina's surface, which may cause scarring and retinal cell loss. Due to a complex grading system and the requirement for trained doctors/ Optometrists to recognise the existence and relevance of numerous tiny characteristics, diagnosing Diabetic Retinopathy (DR) with colour fundus pictures is a challenging and time-consuming task. In this article, we suggest a CNN method for correctly identifying DR from digital fundus images and classifying its severity. We create a network with CNN architecture and data augmentation that can recognise the complex elements needed for the classification task, like micro-aneurysms, exudate, and haemorrhage on the retina, and then deliver a diagnosis automatically and without user input. On the image set as acquired in our previous papers  on Diabetic retinopathy, we train this network using a top-tier graphics processing unit (GPU), and the results are excellent, especially for a challenging classification test. Treatment for diabetic retinopathy is often delayed until it starts to progress to Proliferate DR/ PDR. Comprehensive dilated eye exams are needed more frequently as diabetic retinopathy becomes more severe. People with severe non-proliferative diabetic retinopathy have a high risk of developing PDR and may need a comprehensive dilated eye exam as often as every 2 to 4 months. So, in our paper we have developed such a model where even a thin line of difference between each stages of DR is well distinguished by our model “retina.model” and is 100% reusable with increasing level of cognition with time as the machine tries to learn new patterns.
Keywords: persistently high blood sugar, matrix handling, Diabetes Mellitus, American Optometric Association (AOA), Deep Learning, CNN architecture, Diabetic Retinopathy (DR), Image Classification, retina of the eye, Optometrist, Gaussian filters, Mild DR, Moderate DR, Severe DR, Proliferate DR/ PDR and NO DR.
| DOI: 10.17148/IJARCCE.2022.11914