Case study on detection of diabetic retinopathy with unsupervised deep learning using Retina Care: a web-based application

As per the 10th report of IDF, 1 in 5 diabetes patients experience Diabetic Retinopathy (DR), with 1.3 million adults experiencing DR in 2020 globally. Projections indicated a rise to approximately 3.2 million by 2030. Identifying and addressing DR in its early stages is vital for nations with limited resources, as prompt detection allows for more effective medical interventions and treatment. This work introduces a novel Modified Fuzzy C-Means (MdFCM) algorithm that overcomes the local optima limitation of traditional FCM by refining membership updates for better cluster separation. MdFCM is embedded into the convolutional layer of the Convolutional Neural Network (CNN). In contrast, the fully connected layer is omitted, forming the Fuzzy C-Means Convolutional Neural Network (FCCNN) to achieve more discriminative feature clustering. Combined with the cost-effective MIIRet Cam, the proposed tool achieved 80% accuracy and 88% precision and recall compared to expert grading, positioning it as a reliable preliminary aid. This innovative approach addresses the need for efficient, accessible solutions in resource-limited settings, advancing healthcare interventions for individuals at risk of DR.