Retina Blood Vessels Segmentation Using U-Net

This project aims to segment blood vessels in retina images using a U-Net model. We used a combination of the DRIVE and HRF datasets to enhance the model's generalization capabilities. The dataset was augmented using various image transformations such as HorizontalFlip, VerticalFlip, ElasticTransform, GridDistortion, and OpticalDistortion to further improve performance (augmentation logic can be found in data.py).

Model Overview

• Architecture: U-Net
• Framework: TensorFlow
• Training: The model was trained for 100 epochs on a Kaggle P100 GPU.

Sample Results

Three samples of the model output. The images from left to right are: input image, true mask, and predicted mask.

Dataset

The project combines two retina datasets:

• DRIVE (Digital Retinal Images for Vessel Extraction)
• HRF (High-Resolution Fundus Image Database)

Both datasets were augmented to increase the variety and complexity of training examples.

Augmentation Techniques

We applied the following augmentation techniques:

• Horizontal Flip
• Vertical Flip
• Elastic Transform
• Grid Distortion
• Optical Distortion

This augmentation enhances the model's robustness by exposing it to various transformations of the retina images.