AutoMorph 2022 👀

--Code for AutoMorph: Automated Retinal Vascular Morphology Quantification via a Deep Learning Pipeline.

Project website: https://rmaphoh.github.io/projects/automorph.html

Talks on NIHR Moorfields BRC: https://www.moorfieldsbrc.nihr.ac.uk/news/automorph-tool-to-analyse-retinal-photographs

Please contact [email protected] or [email protected] if you have questions.

Before starting, we summarise the features for three running ways:

• Google Colab (no commands/code, free gpu for 12 hours)
• Configure environment on local/virtual machine (data privacy, code development)
• Docker image (data privacy, no need to configure environment)

Index

• AutoMorph 2022 👀
  • Index
  • Quickstart with Colab
  • Install instruction for local/virtual machine
    • Requirements
    • Package installation
    • Running
  • Docker usage
  • Common questions
    • Memory/ram error
    • Invalid results
  • Components
  • Citation

 

Pixel resolution

The units for vessel average width, disc/cup height and width, and calibre metrics are defined as microns. For it, we need to organise a resolution_information.csv which includes the pixel resolution information, which can be queried in FDA and Dicom files. Alternatively, some people use approximate value for every images, e.g., 0.008 for Topcon 3D-OCT.

If you don't use these features or care their units, you can just run following command after putting all images in the folder of ./images

python generate_resolution.py

Quickstart with Colab

Use the Google Colab and a free Tesla T4 gpu

Colab link click

Install instruction for local/virtual machine

Requirements

1. Linux is preferred. For windows, install MinGW-w64 for using commands below to set enviroment.
2. Anaconda or miniconda installed.
3. python=3.6, cudatoolkit=11.0, torch=1.7, etc. (installation steps below)
4. GPU is essential.

Package installation

Step 1: create virtual environment:

conda update conda
conda create -n automorph python=3.6 -y
conda activate automorph

Step 2: install pytorch 1.7 and cudatoolkit 11.0

conda install pytorch==1.7.0 torchvision==0.8.0 torchaudio==0.7.0 cudatoolkit=11.0 -c pytorch -y

Step 3: install other packages:

pip install --ignore-installed certifi
pip install -r requirement.txt
pip install efficientnet_pytorch

Running

Activate virtual environment and clone the code.

conda activate automorph
git clone https://github.com/rmaphoh/AutoMorph.git

Put the images in folder 'images' and

sh run.sh

Please not that resolution_information.csv includes the resolution for image, i.e., size for each pixel. Please prepare it for the customised data in the same format.

 

Docker usage

Zero experience in Docker? No worries.

First, clone the github to <path/of/AutoMorph, e.g., /home/AutoMorph> and put the images in AutoMorph/images

git clone https://github.com/rmaphoh/AutoMorph.git

Then, pull our docker image and run the tool.

docker pull yukundocker/image_automorph
docker run  -v <path/of/AutoMorph>:/root/AutoMorph -ti --runtime=nvidia -e NVIDIA_DRIVER_CAPABILITIES=compute,utility -e NVIDIA_VISIBLE_DEVICES=all yukundocker/image_automorph
source /root/set_up.sh

Common questions

Memory/ram error

We use Tesla T4 (16Gb) and 32vCPUs (120Gb). When you meet memory/ram issue in running, try to decrease batch size:

• ./M1_Retinal_Image_quality_EyePACS/test_outside.sh -b=64 to smaller, e.g., 32 or 16.
• ./M2_Artery_vein/test_outside.sh --batch-size=8 to smaller
• ./M2_lwnet_disc_cup/test_outside.sh --batchsize=8 to smaller

Invalid results

In csv files, invalid values (e.g., optic disc segmentation failure) are indicated with -1.

Components

1. Vessel segmentation BF-Net

2. Image pre-processing EyeQ

3. Optic disc segmentation lwnet

4. Feature measurement retipy

Citation

@article{zhou2022automorph,
  title={AutoMorph: Automated Retinal Vascular Morphology Quantification Via a Deep Learning Pipeline},
  author={Zhou, Yukun and Wagner, Siegfried K and Chia, Mark A and Zhao, An and Xu, Moucheng and Struyven, Robbert and Alexander, Daniel C and Keane, Pearse A and others},
  journal={Translational vision science \& technology},
  volume={11},
  number={7},
  pages={12--12},
  year={2022},
  publisher={The Association for Research in Vision and Ophthalmology}
}