An automated unsupervised deep learning-based approach for diabetic retinopathy detection.

As per the International Diabetes Federation (IDF) report, 35-60% of people suffering from diabetic retinopathy (DR) have a history of diabetes. DR is one of the primary reasons for blindness and visual impairment worldwide among adults aged 24-74 years. Therefore, this research aims to develop an automated technique for the detection of retinal abnormalities associated with DR, such as microaneurysm. Unsupervised learning has a high potential for data classification. The proposed work accomplishes the following objectives. (a) k-means and fuzzy clustering method is discussed, and the objective function is revised to offer the modified version named modified fuzzy clustering method (MdFCM). (b) A modified convolutional neural network is proposed to consolidate the MdFCM and features for better outcomes. (c) The results are compared on three diverse datasets, DIARETDB1, APTOS, and Liverpool, with the fuzzy clustering method, deep embedded clustering, and k-means for generalizability. To the best of our knowledge, the proposed algorithm is the first to detect DR using a hybrid approach of unsupervised and deep learning methodology. The proposed system achieved an improved accuracy rate of 98.6%. The results show that our proposed method outperforms the state-of-the-art algorithm. We intend to design a tool using the proposed system for diabetic retinopathy detection at an early stage. Complete system flow architecture of diabetes retinopathy detection using unsupervised deep learning approach.