varshaak/retinal-vessel-segmentation

Retinal Vessel Segmentation

An interactive Streamlit app for exploring retinal vessel segmentation and exploring how algorithm choice and parameter tuning affect clinical measurements. Live demo: https://huggingface.co/spaces/varshaak/retinal-vessel-segmentation

## What the app does This app demonstrates a full image segmentation pipeline applied to retinal fundus images used in clinical screenings for diabetic retinopathy, hypertension and glaucoma.

### Pipeline steps:

1. 1.Green channel extraction- hemoglobin absorbs green light, hence vessels have highest contrast in this channel.
2. 2.CLAHE enhancement- locally equalises contrast to handle uneven illumination across the retina.
3. 3.Segmentation- two methods available:
4. 4.Otsu thresholding- global intensity-based vessel detection
5. 5.Watershed- marker-based segmentation that handles touching vessels
6. 6.Clinical metrics- vessel dencity (%), vessel count, mean vessel width

### Clinical relevance:

• —Vessel density is a measurable biomarker for diabtic retinopathy
• —Vessel calibre changes with hypertension and cardiovascular risk
• —Reproducible, automated segmentation is essential for multi-site clinical trials

## Sample images Sample images are sourced from the DRIVE dataset which is the standard benchmark dataset in retinal vessel segmentation research.

Screenshots

Full App Overview

[image]

Pipeline Steps

[image]

Otsu Thresholding

[image]

Watershed Segmentation

[image]

Diagnostic View

[image]

## Running locally

1. 1.Download or clone this Space from Hugging Face
2. 2.Install dependencies: pip install -r requirements.txt
3. 3.Run the app: streamlit run app.py Python 3.9+ recommended.

Project structure

retinal-segmentation/
├── app.py              # Streamlit interface
├── processing.py       # Image analysis pipeline
├── requirements.txt    # Dependencies
├── README.md
└── sample_images/      # Bundled DRIVE sample images

Dependencies

• —Streamlit - web interface
• —scilit-image - image processing
• —OpenCV - image I/O and resizing
• —SciPy - distance transforms and morphology
• —Matplotlib - visualisation