DhruvPremani1017/Diabetic-Retinopathy

This project uses a U-Net model to segment key diabetic retinopathy lesions from retinal images in the IDRiD dataset. It detects microaneurysms, haemorrhages, exudates, and the optic disc to support early DR diagnosis and aid clinicians in low-resource settings. # Diabetic Retinopathy Lesion Segmentation Using the IDRiD Dataset This project implements a deep learning-based segmentation model to identify diabetic retinopathy (DR) lesions from retinal fundus images using the U-Net architecture. The model is designed to segment five key components associated with DR: - Microaneurysms (MA) - Haemorrhages (HE) - Hard Exudates (EX) - Soft Exudates (SE) - Optic Disc (OD) ## ⚙️ Technical Approach A U-Net-based convolutional neural network is used for pixel-wise segmentation of lesions from fundus images. The training is conducted on the IDRiD dataset, which offers high-resolution images with precise pixel-level annotations. **Note:** This project is still a work in progress. The repository will be updated continuously. ## 📁 Dataset Info - **Dataset Name:** Indian Diabetic Retinopathy Image Dataset (IDRiD) - **Access Link:** IEEE Dataport – IDRiD Dataset - **Challenge:** Diabetic Retinopathy Segmentation and Grading Challenge – ISBI 2018 > IDRiD is the first diabetic retinopathy dataset representative of the Indian population. It contains pixel-level annotations for both pathological lesions and normal anatomical structures, making it ideal for developing AI-based diagnostic tools. --- Feel free to explore, fork, or contribute to this repository to improve the model, experiment with training techniques, or evaluate new strategies for lesion detection.