The global prevalence of diabetes is increasing at a rapid rate, with projections indicating that the number of people with diabetes will surpass 1.31 billion by 2050. Diabetic retinopathy (DR) is a common vision complication that arises from diabetes and is a leading cause of preventable blindness worldwide. The magnitude of this disease's impact calls for modern and more efficient solutions for DR detection; early detection of the disease can lessen the risk of vision loss substantially for those inflicted with it: with quick intervention, the risk of severe vision loss can be reduced by as much as 90%. In this paper, we utilize and finetune the Segment Anything Model(SAM) to segment images for feature extraction in the Indian Diabetic Retinopathy Image Dataset (IDRiD), an open-access dataset of retinal images structured for retinopathy screening research. We then employed a Convolutional Neural Network (CNN) architecture, specifically a Keras model, to classify the key features within these segmented images and grade the severity of DR using ETDRS Classification, a semi-qualitative grading scale that assigns a severity score of 0 to 4 to patients. Our successful implementation of SAM in conjunction with CNN architecture demonstrates the promise of this approach in diabetic retinopathy and disease classification as a whole.