Keywords: Image Synthesis, Segmentation, Generative Adversarial Networks, Diabetic Retinopathy
Abstract: Automatic segmentation of retina lesions have been a long standing and challenging task for learning based models, mostly due to the lack of available and accurate lesion segmentation datasets. In this paper, we propose a two-step process for generating photo-realistic fundus images conditioned on synthetic "ground truth" semantic labels, and demonstrate its potential for further downstream tasks, such as, but not limited to; automated grading of diabetic retinopathy, dataset balancing, creating image examples for trainee ophthalmologists, etc.
Paper Type: both
Primary Subject Area: Image Synthesis
Secondary Subject Area: Segmentation
Paper Status: original work, not submitted yet
Source Code Url: https://github.com/sonjoonho/semantic-retina-generation
Data Set Url: IDRiD https://ieee-dataport.org/open-access/indian-diabetic-retinopathy-image-dataset-idrid, FGADR https://csyizhou.github.io/FGADR/, Synthetic https://drive.google.com/drive/folders/19czqBOKrJTinVt_Zz8mmdb-r2epvl4K6?usp=sharing
Registration: I acknowledge that publication of this at MIDL and in the proceedings requires at least one of the authors to register and present the work during the conference.
Authorship: I confirm that I am the author of this work and that it has not been submitted to another publication before.
TL;DR: We use synthetic retina images to improve lesion segmentation performance.