<p>People who have diabetes are prone to the progress of Diabetic Retinopathy (DR) and Diabetic macular Edema (DME). DR is related with diabetes, in human eye it can able to cause permanent vision impairment. Therefore, it is necessary to predict DR in early stages. All stages of DR has a chance to have DME and it affects 1 in 15 patients who suffers with diabetes. CNN models with fundus images have been used in grading of DR for faster screening. The present research uses Modified Block level Attention Group-CNN (MBAttG-CNN) for the classification of DR and DME which uses IDRiD dataset. CNN highly used in the area of image recognition because of the enhanced performance. CNN model use Convolutional layers in order to learn feature map and block level attention mechanism is used for the computation of new feature map. In the proposed model, CNN enhanced with block level attention mechanism helps in formulating the solution which assists the detection of DR grading and DME in early stages of the disease. The proposed method helps to detect the grading of DR and DME which assists to find the severity. The accuracy rate of 98.07% is achieved on IDRiD dataset that establishes the superior performance of the proposed method.</p>