A Novel Approach for the Early Recognition of Diabetic Retinopathy using Machine Learning

Diabetic Retinopathy (DR) infection is a foremost cause of sightlessness and most chronic ocular diseases provoked through the complications of diabetes which influences eye that arises when retinal blood vessels become swelled and fluid leaks, sometimes blood vessels become a block and preventing blood from passing through, that outcomes in atypical unwanted blood vessels cultivate on the retina. All of these changes eventually lead to vision loss or even blindness, if left untreated. Early recognition is very necessary for successful treatment but unluckily, accurate identification of DR is a challenging task and also requires the Ophthalmologists to manually analyzing the fundus image. Presently, the DR infection recognition system is extremely arduous and prolonged. For eradicating these problems, necessarily require an automatic detection system, therefore the ambition of this effort is to categorize the unusual eye blood vessels, exudates dots, cotton wool marks from fundus images. The ocular image is pre-processed via histogram equalization procedure then improved image is segmented using k-mean clustering. SVM and Random Forest classification algorithm plays an extremely significant job for the categorization of usual and unusual areas of the ocular image and also helpful for plummeting human error that diminishes false recognition and acquires a high precision rate. The recognition rate of the Random Forest classifier is 96.62% that expresses a great and stable outcome as compared to the SVM. The dataset contains a total no. of 89 color images out of which 84 have gentle non-proliferative signs (Microaneurysms) of the DR infection, and 5 pictures are taken as a usual image that does not hold any indication of the DR symptoms.