DiscTorsionAnalyzer 👁️‍🗨️

Software to automatically measure optic disc torsion in colour fundus photographs of the retina

Table of contents

• Description
• Quick start
• Status
• Bugs and feature requests
• Contributing
• Creators
• Copyright and license

Description

Optic disc torsion refers to the deviation of the longest axis of the optic disc from its perpendicular meridian. It is a measure of how much the optic disc has rotated from its normal orientation relative to the eye's vertical axis. Measuring optic disc torsion is important because it can provide valuable insights into various ocular and systemic conditions.

This software uses deep learning-based segmentation to estimate en ellipse around the optic and locate the fovea. It then rotates the image along the centrepoints of these two key structures and measures the angle between the vertical line centred on the optic disc and the major axis of the ellipse. The resulting value is either positive (indicating inferotemporal torsion) or negative (indicating supranasal torsion).

Ovality is also measured, defined as the major axis length / minor axis length. A ovality value larger than 1.3 is generally considered optic disc tilt.

Manuscript currently in preparation. This measure is currently NOT CLINICALLY VALIDATED. For research purpose only.

Quick start

Requirements

1. Linux OS
2. Miniconda
3. GPU

Installation

Step 1: create virtual environment

conda update conda
conda create -n DiscTorsionAnalyzer python=3.12 -y

Step 2: activate virtual environment and clone repo

conda activate DiscTorsionAnalyzer
git clone https://github.com/samuel-gibbon/DiscTorsionAnalyzer.git
cd DiscTorsionAnalyzer

Step 3: install torch

pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118

If this step fails, go to https://pytorch.org/get-started/locally/ and follow installation instructions.

Step 4: install other packages

pip install -r requirements.txt

Running

Open the config.ini file and set pathIn and pathOut
You can also disable writeImages and/or writeResults if needed.

python main.py

Results

One processed image is stored per input image, and numerical results are output in a results.csv file. Everything is written to pathOut. Example below on an image from the publicly available IDRiD dataset.

Filename	Ovality	Torsion
IDRID_28.png	0.89	12.12

Status

Software is currently working for colour fundus images of any resolution and format (e.g., PNG, JPG).

Bugs and feature requests

Have a bug or a feature request? Please first read the issue guidelines and search for existing and closed issues. If your problem or idea is not addressed yet, please open a new issue.

Contributing

Please read through our contributing guidelines. Included are directions for opening issues, coding standards, and notes on development.

Creators

Creator 1

Created by Samuel Gibbon and Borja Marin

• https://github.com/samuel-gibbon/DiscTorsionAnalyzer.git

Copyright and license

Code and documentation copyright 2023-2024 the authors. Code released under the MIT License.