DeepSeeNet: A Python repository from codevisioner

DeepSeeNet is a high-performance deep learning framework for grading of color fundus photographs using the AREDS simplified severity scale. For more details, please see https://ncbi-nlp.github.io/DeepSeeNet/.

Getting Started with DeepSeeNet

These instructions will get you a copy of the project up and run on your local machine for development and testing purposes. The package should successfully install on Linux.

Installing

Prerequisites

python =3.6
tensorflow >=1.6.0
keras =2.2.4
Linux

Tensorflow can be downloaded from https://www.tensorflow.org.

Installing from source

Download the source code from GitHub: git clone https://github.com/ncbi-nlp/DeepSeeNet.git
Change to the directory of DeepSeeNet
Install required packages: pip install -r requirements.txt
Add the code directory to PYTHONPATH: export PYTHONPATH=.:$PYTHONPATH

Using DeepSeeNet for grading simplified scores

The easiest way is to run the following command

$ python examples/predict_simplified_score.py data/left_eye.jpg data/right_eye.jpg
...
Downloading data from https://github.com/ncbi-nlp/DeepSeeNet/releases/download/0.1/drusen_model.h5
INFO:root:Loading the model: /tmp/.keras/datasets/drusen_model.h5
Downloading data from https://github.com/ncbi-nlp/DeepSeeNet/releases/download/0.1/pigment_model.h5
INFO:root:Loading the model: /tmp/.keras/datasets/pigment_model.h5
Downloading data from https://github.com/ncbi-nlp/DeepSeeNet/releases/download/0.1/advanced_amd_model.h5
INFO:root:Loading the model: /tmp/.keras/datasets/advanced_amd_model.h5
...
INFO:root:Processing: data/left_eye.jpg
INFO:root:Processing: data/right_eye.jpg
...
INFO:root:Risk factors: {'pigment': (0, 0), 'advanced_amd': (0, 0), 'drusen': (2, 2)}
The simplified score: 2

The script will

Download the models from the DeepSeeNet repository
Predict the simplified score based on the sample left and right eyes

More options (e.g., setting the models) can be obtained by running

$ python examples/predict_simplified_score.py --help

Pre-trained DeepSeeNet models

Besides grading the simplified score, we also provide individual risk factor models. For example

$ python examples/predict_drusen.py data/left_eye.jpg
...
INFO:root:Loading the model: /tmp/.keras/datasets/drusen_model.h5
...
INFO:root:Processing: data/left_eye.jpg
...
The drusen score: [[0.21020733 0.2953384  0.49445423]]
The drusen size: large

Here, we provide the following pre-trained models:

drusen size: non/small, intermediate, large
pigmentary abnormalities: no, yes
late AMD: no, yes
geographic atrophy (GA): no, yes
central GA: no, yes

Training DeepSeeNet model

You can train the individual risk factor model too. For example

$ python examples/train.py data/pigment_label_sample.csv data/pigment_best_model.h5
...
Epoch 1/100
2/2 [==============================] - 27s 14s/step - loss: 1.0103 - acc: 0.5148...
...
early stopping

The program will read images and labels from a CSV file, train the model, and save the latest best model according to the val_acc.

Acknowledgments

This work was supported by the Intramural Research Programs of the National Institutes of Health, National Library of Medicine and National Eye Institute.

Citing DeepSeeNet

If you're running the DeepSeeNet framework, please cite:

Peng Y*, Dharssi S*, Chen Q, Keenan T, Agron E, Wong W, Chew E, Lu Z. DeepSeeNet: A deep learning model for automated classification of patientbased age-related macular degeneration severity from color fundus photographs. Ophthalmology. 2019. 126(4), 565-575.
Keenan T*, Dharssi S*, Peng Y*, Chen Q, Agron E, Wong W, Lu Z, Chew E. A deep learning approach for automated detection of geographic atrophy from color fundus photographs. Ophthalmology. 2019 (Accepted).

NCBI's Disclaimer

This tool shows the results of research conducted in the Computational Biology Branch, NCBI.

The information produced on this website is not intended for direct diagnostic use or medical decision-making without review and oversight by a clinical professional. Individuals should not change their health behavior solely on the basis of information produced on this website. NIH does not independently verify the validity or utility of the information produced by this tool. If you have questions about the information produced on this website, please see a health care professional.

More information about NCBI's disclaimer policy is available.

About text mining group.

For Research Use Only

The performance characteristics of this product have not been evaluated by the Food and Drug Administration and is not intended for commercial use or purposes beyond research use only.

codevisioner/DeepSeeNet