Diabetic Retinopathy Image Classification using Quasi Cross UNet Discrete Transform with Gold Rush Optimizer

The diagnosis of diabetic retinopathy faces three main challenges which include inconsistent image clarity and weak detection of small visual impairments and little ability to share information between different sets of data. The model suffers from two issues which limit its practical use: high expenses for operations as well as low decision accuracy because of complex tissue dynamics. Using the MESSIDOR dataset as input, this paper illustrates the application of the Quasi Cross UNet Discrete Transform with Gold Rush Optimizer (QCUNet-DT-GRO) algorithm for diabetic retinopathy picture categorization. The Unet Swin Transformer (UST) approach is used to precisely segment areas afflicted by diabetic retinopathy, while the Quasi-Cross Bilateral Filtering (QCBF) methodology improves image quality. The One-Dimensional Discrete Transform (ODDT) is used to extract features. The Gold Rush Optimizer (GdRO) does optimization, while the Gates-Controlled Deep Unfolding Network (GCDUNet) handles classification. The system detects diabetic retinopathy along with distinguishing between normal and diabetic retina structures. According to experimental results, QCUNet-DT-GRO performs better than current techniques, with 99.9% accuracy and 99.8% sensitivity. This method is shown to be a viable substitute for the manual diagnostic processes that are now in use, enabling robotic diagnosis systems in subsequent applications.