Determination of Non-Proliferative and Proliferative Diabetic Retinopathy through Fundoscopy using Principal Component Analysis

Image processing can be one of numerous ways that can be utilized in recognizing illnesses particularly in the retina of the eye as it isn't effectively available for examination except the eye is to be dilated and can give a clearer image of it with the help of a lens. The characterization of existing models was constrained to determining Non-Proliferative Diabetic Retinopathy (NPDR) and excluding Proliferative Diabetic Retinopathy (PDR). In this research, a diabetic retinopathy determination system was created on Raspberry Pi 3 Model B by performing Pre-processing methods on images and through Principal Component Analysis and Support Vector Machine for classification on the OpenCV software. The system is also able to characterize the retina as normal, NPDR, or PDR. The system was tested and observed for the purpose of obtaining clear images and its accuracy through the Confusion Matrix. Combing Principal Component Analysis together with Support Vector Machine provides an accurate result of 86.67% with some errors due to having some minor similarities in the retina when processed such as camera flash must be mistaken as cotton wool spots, haemorrhages, and micro aneurysms. Providing a larger pool of database will effectively increase the accuracy rate of the system.