This repository contains code for segmenting retinal fundus images using deep learning techniques. The segmentation process involves identifying and isolating specific structures within the retinal images, such as blood vessels, optic discs, and lesions.

Currently, the main focus of segmentation is on identifying the lesions. The segmentation is performed using an U-Net model, trained on a dataset of retinal fundus images and their corresponding masks.

The segmentation process is implemented in a Jupyter notebook (segmentation.ipynb), which includes data preprocessing, model training, and visualization of results.
The loss function used for training the model is a combination of Binary Cross-Entropy (BCE) and Dice Loss, which helps in improving the accuracy of segmentation.

• The dataset used: e-ophtha-EX Dataset

Retinal Fundus Images	Corresponding Annotations

• Recall: 82.87%
• Precision: 76.14%
• F1-Score: 79.34%

• Remove optic disc from the images to reduce false positives.
• Experiment with different architectures and hyperparameters to improve segmentation accuracy.
• Post-processing techniques to refine the segmented masks.

Suggestions and contributions are welcome! 🙂