Publication

Abstract : Among people with diabetes, diabetic maculopathy, a dangerous consequence of diabetic retinopathy is the primary cause of visual impairment. Effective treatment and management depend on early and precise identification. In this research, datasets from Kaggle, DRIVE, CHASE, STARE and DIARETDB1 is employed and a Convolutional Neural Network (CNN) based method is used for recognizing the presence of diabetic maculopathy. Multiple convolutional layers make up the model architecture, which is then followed by pooling and fully connected layers. A SoftMax method is then used to categorize the data and is extensively tuned for hyperparameters. By incorporating transfer learning from pre-trained models, the performance is further improved and a significant reduction in training period and processing resources is required. Using a variety of datasets guarantees a thorough assessment of the model, improving its dependability and generalizability. The model's capacity to retain excellent performance across several data sources is confirmed by independent testing and cross-validation on each dataset. By facilitating prompt medical intervention, the suggested CNN model might possibly improve patient outcomes by aiding in the early detection of diabetic maculopathy. With a sensitivity of 90%, specificity of 98%, precision of 97.8%, accuracy of 94%, and F1 score of 93.8%, our model performed well. These measurements highlight the promise of deep learning approaches in improving medical imaging diagnostics by demonstrating the model's resilience and efficacy in diagnosing diabetic maculopathy. Accuracy obtained using completed Local Binary Pattern (CLBP) and Deep Learning-Based Framework for Automatic Diabetic Retinopathy Detection is found to be 90% and 91 %, whereas the proposed method offers an accuracy of 94% which is better and optimum.