Enhanced Diabetic Retinopathyanalysis: Unet-Based Lesion Segmentation Coupled with Mobilenetv2 for Feature Extraction and Efficientnetb0 for Classification

In this research, we address the critical issue of precise Diabetic Retinopathy (DR) diagnosis, a condition of ten leading severe vision impairment or blindness in diabetic patients. Leveraging advanced deep learning models and innovative image processing techniques, our study focuses on accurate retinal image segmentation using a UNet model. This segmentation method delineates lesions and retinal structures effectively. Subsequently, Gabor filters are applied for intricate texture pattern extraction, indicative of diverse retinopathy stages. Integrating MobileNetV2 for feature extraction and EfficientNetB0 for multi-class classification significantly enhances the diagnostic accuracy. Our developed system exhibits a promising 91.2% test accuracy, showcasing its potential in DR diagnosis. While challenges related to varying severity levels persist, our robust framework lays the groundwork for future refinements. By amalgamating sophisticated image segmentation, feature extraction, and classification techniques, our system provides a solid foundation for accurate and timely DR assessment. With continuous enhancements, including the incorporation of more extensive and diverse datasets, our approach holds the promise to revolutionize DR diagnostics. The integration of cutting-edge technology into medical practices underscores the transformative impact of artificial intelligence in the realm of ophthalmology, promising improved patient outcomes.