Deep Learning based Automated Diabetic Retinopathy Classification

One of the serious side effects of diabetes is diabetic retinopathy (DR), which causes blindness and progressive vision loss. early and precise detection and assessment of DR severity is critical for treatment. Retinal scans are frequently used for subjective analysis by the medical experts, which is time-consuming, and error-prone manual diagnosis. To overcome these challenges, this study proposed an automated DR classification framework by employing advanced image processing techniques and combined with NASNet-Large model, which is an effective convolutional neural network (CNN) architecture discovered via neural architecture search. The implemented methodology includes green-channel extraction, and enhancement by contrast enhancement (CLAHE) and followed by segmentation for retinal features based on morphological transformations. The NASNet-Large model is then trained and validated on standardized retinal datasets (Messidor-1), achieving accurate classification across multiple DR severity levels. The Experimental results obtained shows good performance with an accuracy of 92.2%, and highlighting the NASNet-Large’s capability to in effectively recognizing the retinal features. The proposed system significantly enhances the automated DR screening, and offers robust diagnostic assistance by reducing the ophthalmologists' workload and impacts better patient outcomes.