Detection Of Diabetic Retinopathy Using Machine Learning & Image Processing

Abstract

This research proposes a comprehensive framework for diabetic retinopathy detection, encompassing five key stages: image passing, data preprocessing, data augmentation, feature extraction, and classification. The initiation is done with the acquisition of the retinal images that undergo the thorough data preprocessing procedures that are geared towards the enhancement of the quality and clarity. Afterward, the data augmentation process is initiated to diversify the dataset, augmenting its size and ensuring robust model training. Feature extraction follows, utilizing advanced algorithms to identify distinctive characteristics associated with diabetic retinopathy. Each stage is accompanied by rigorous evaluation, yielding results that highlight the effectiveness of the proposed methods. The final classification stage employs machine learning models, particularly deep learning techniques namely VGG16, to categorize retinal images based on identified features. The comprehensive nature of this framework, from image acquisition to classification, contributes to the development of an automated system for diabetic retinopathy detection. Results demonstrate the system's efficacy in accurately identifying and classifying diabetic retinopathy, showcasing its potential for widespread clinical application and enhancing early diagnosis for improved patient outcomes. The proposed pre-trained VGG16 model confirmed better accuracy by using the Adam optimizer in DR detection when applied with the approach of transfer learning (TL). The model is not overfitted and it provides the more accurate results.