Exploring the future of diabetic retinopathy classification: a roadmap from deep learning to quantum computing

Diabetic Retinopathy (DR) is one of the leading causes of blindness worldwide. The current technological advancements, such as machine learning and deep learning, significantly enhance the detection of DR. However, the further development of early and accurate diagnostic procedures remains essential to preserve human vision. To support this, the development of deep learning models applied to retinal image analysis has substantially improved the detection and classification accuracy of DR. In particular, transfer learning with pre-trained networks drastically tappers the need for massive, labelled datasets by eliminating the high computational requirements. Quantum and quantum-inspired approaches exhibit theoretical promise for selected machine learning tasks. However, their application to diabetic retinopathy remains largely experimental and is predominantly confined to evaluations conducted on classical simulation platforms. By keeping the above points as a roadmap, this review provides an structured overview of diagnostic and classification methods for DR. Moreover, this scoping review procedure has been performed through machine learning, deep learning, transfer learning, and quantum computing by analyzing their pros and cons. The outcome of the review emphasizes that while classical deep learning and transfer learning demonstrate clinically validated performance, quantum-inspired approaches remain in early research stages and require substantial hardware development and clinical validation before practical deployment. It also identifies future research avenues focused on developing hybrid models that integrate complementary strengths of these technologies, as well as exploring quantum-inspired algorithms tailored for DR image analysis.