A Deep Learning Approach for Classifying Retinal Diseases and Diabetic Retinopathy Stages to Enhance Early Detection in Diagnosis

Diabetic Retinopathy (DR) is a common consequence of diabetes mellitus, resulting in retinal lesions that can adversely affect vision. If undetected and mistreated, diabetic retinopathy can ultimately result in blindness. This study concentrates on creating an enhanced model for the precise classification of retinal illnesses, particularly Diabetic Retinopathy. The suggested model increases the precision and effectiveness of disease classification by utilising deep learning methods and attention mechanisms. The study analyses the performance of many pre-trained models, such as ResNet152, EfficientNetB7, and MobileNetV3Large, using a varied dataset of retinal pictures. The ResNet152 model, enhanced with the Spatial Attention Module, achieved the best results in retinal disease classification, with an accuracy of 87.65(%), precision of 89.88(%), recall of 90.69(%), and F1-score of 88.58(%). This performance surpasses other models and demonstrating the effectiveness of attention modules in improving classification accuracy and robustness. The study also highlights the competitive performance of the suggested model by contrasting it with earlier studies in the field. All things considered, this research advances the classification of retinal diseases and may improve patient treatment and ophthalmology diagnostic precision.