Diabetic Retinopathy Classification Using Pyramidal Convolution Shuffle Attention Neural Network

Diabetic retinopathy (DR) is the most dangerous consequences of diabetes that cause lifelong blindness. It has become more widespread in recent years. Furthermore, diabetic patients' retinal blood vessels are damaged by DR. Consequently, several DL with AI methods have been put out to automatically detect anomalies in DR, distinct phases from photographs of the retina. This study introduces DRC-PCS-ANN, a novel neural network architecture based on pyramidal convolution and shuffle attention, for effective diabetic retinopathy classification. As a first step, input images were acquired from the MESSIDOR database. Following acquisition, noise is removed from the images through pre-processing using Data-Adaptive Gaussian Average Filtering (DAGAF). In the subsequent step, statistical features like Mean, Kurtosis, Variance, and Entropy are extracted using the One-Dimensional Quantum Integer Wavelet S-Transform. Finally, the extracted features are fed into PCSANN for effective classification of Diabetic Retinopathy into five categories: No DR, Mild DR, Moderate DR, Severe DR, and Proliferative DR. Then, performance of DRC-PCS-ANN examined utilizing performance metrics like accuracy, precision, F1-score. The DRC-PCS-ANN method attains 21.28%, 21.52% and 20.34% higher accuracy, 23.29%, 23.83% and 21.72% higher precision and compared with existing DR classifiers such as DR Classification depend on Multipath CNN with ML classifiers (DRC-MCNN-MLC), Automated DR screening for Primary Care Settings utilizing DL (ADR-PCS-DL) and Automatic DR diagnosis utilizing Adaptive Fine-Tuned CNN (ADR-AFT-CNN) respectively.