Code Repository for High-Fidelity Diabetic Retina Fundus Image Synthesis from Freestyle Lesion Maps.
RetinaGAN a two-step process for generating photo-realistic retinal Fundus images based on artificially generated or free-hand drawn semantic lesion maps.
StyleGAN is modified to be conditional in to synthesize pathological lesion maps based on a specified DR grade (i.e., grades 0 to 4). The DR Grades are defined by the International Clinical Diabetic Retinopathy (ICDR) disease severity scale; no apparent retinopathy, {mild, moderate, severe} Non-Proliferative Diabetic Retinopathy (NPDR), and Proliferative Diabetic Retinopathy (PDR). The output of the network is a binary image with seven channels instead of class colors to avoid ambiguity.
The generated label maps are then passed through SPADE, an image-to-image translation network, to turn them into photo-realistic retina fundus images. The input to the network are one-hot encoded labels.
Download model checkpoints (see here for details) and run the model via Streamlit. Start the app via streamlit run web_demo.py.
Example retina Fundus images synthesised from Conditional StyleGAN generated lesion maps. Top row: synthetically generated lesion maps based on DR grade by Conditional StyleGAN. Other rows: synthetic Fundus images generated by SPADE. Images are generated sequentially with random seed and are not cherry picked.
| grade 0 | grade 1 | grade 2 | grade 3 | grade 4 |
|---|---|---|---|---|
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
If you find this work useful for your research, give us a kudos by citing:
@misc{retinagan,
author={Benjamin Hou and Amir Alansary and Daniel Rueckert and Bernhard Kainz},
title={High-Fidelity Diabetic Retina Fundus Image Synthesis from Freestyle Lesion Maps},
year={2022},
}