Automatic diabetic retinopathy classification using probabilistic spiking heterogeneous dynamic graph network with hyperbolic sine optimizer

Diabetic retinopathy (DR) is a major cause of vision loss, requiring accurate and early detection for effective treatment. However, existing automated methods often suffer from limited generalization, weak feature representation, and inefficient optimization. To overcome these problems, the current paper has proposed a new Probabilistic Spiking Heterogeneous Dynamic Graph Network with Hyperbolic Sine Optimizer (PSHDGNet-HSO) for automatic detection of DR from fundus images. The architecture combines graph-based relation learning with probabilistic spiking neuron calculation to understand intricate interactions between different regions of retina, and the Hyperbolic Sine Optimizer guarantees optimal parameter tuning. The algorithm tests its performance on the APTOS data set through an organized workflow which includes three main activities of pre-processing and segmentation and feature extraction and classification. Experimentation shows improved results over previous works, attaining higher levels of accuracy and precision.