Segmentation of hard exudates in fundus images to detect diabetic retinopathy using modified U-NET

Diabetic retinopathy (DR) is a severe complication of diabetes, leading to potential vision loss due to damage to the retinal blood vessels. Hard exudates, which are visible lesions in fundus images, serve as critical indicators for diagnosing and monitoring DR. This study introduces a Modified U-Net architecture designed to improve the segmentation of hard exudates, thereby enhancing the detection and management of diabetic retinopathy.The U-Net model, renowned for its effectiveness in biomedical image segmentation, is adapted with several enhancements to better address the complexities of fundus images. These modifications include advanced feature extraction techniques, integration of attention mechanisms to focus on significant areas, and refined post-processing methods. These improvements aim to increase the accuracy and reliability of hard exudate segmentation.The Modified U-Net is evaluated on a dataset of fundus images with annotated hard exudates, using performance metrics such as accuracy, precision, recall, and the Dice coefficient. The results reveal that the Modified U-Net significantly outperforms traditional U-Net models and other contemporary segmentation methods. This enhanced model not only achieves higher accuracy in detecting and segmenting hard exudates but also improves the overall sensitivity and specificity.