The present study aims to study the application of unsupervised Deep Belief Networks to the clustering of diabetic retinopathy images to aid in disease severity classification. A dataset of 2300 images is obtained from online databases and Kaggle, each of which is labeled with the severity of the disease. The DBNs are trained to extract features from the images to be used for clustering process. K-Means clustering, Gaussian Mixture Models and Hierarchical Clustering are performed on the features extracted. Metrics like Silhouette Score, Calinski-Harabasz Index, Davies-Bouldin Index and Dunn Index are used to evaluate the performance of each of the clustering techniques in terms of quality and effectiveness. The results show that the K-Means clustering is the most effective technique for accurately clustering the images with respect to disease severity. Higher scores are obtained in this clustering method in terms of things like cluster cohesion, separation, and compactness. The Silhouette Score is 0.75, Calinski-Harabasz Index is 452.89, Davies-Bouldin Index is 0.42 and Dunn Index is 0.76 in the case of K-Means. In the case of GMM, the Silhouette Score is slightly lower at 0.68, Calinski-Harabasz Index is 387.62, Davies-Bouldin Index is 0.49 and Dunn Index is 0.68. In the case of HC, the Silhouette Score is 0.62, Calinski-Harabasz Index is 318.47, Davies-Bouldin Index is 0.57 and Dunn Index is 0.61. Thus, K-Means clustering is suitable for classifying the images with respect to disease severity.