Improving cross-domain diabetic retinopathy lesions segmentation based on CycleGAN augmentation

Yun Zhang; Hao Zhou; Zhi Xie; Yao He

doi:10.1117/12.2636505

10 June 2022 Improving cross-domain diabetic retinopathy lesions segmentation based on CycleGAN augmentation

Yun Zhang, Hao Zhou, Zhi Xie, Yao He

Proceedings Volume 12179, Second International Conference on Medical Imaging and Additive Manufacturing (ICMIAM 2022); 121791B (2022) https://doi.org/10.1117/12.2636505
Event: Second International Conference on Medical Imaging and Additive Manufacturing (ICMIAM 2022), 2022, Xiamen, China

Abstract

Diabetic retinopathy (DR) severity grade depends on lesion types. Automatic lesion segmentation of DR on fundus image plays a key role in the diagnosis of DR. It is increasingly common that a model is trained by the images from different sources. While a model trained on the source domain is transferred to another (target domain), the performance of the model generally decreases. In this paper, a novel method was proposed for cross-domain segmentation of DR lesions by applying cycle-consistent adversarial networks (CycleGAN) and an improved Xception-based UNet named AttXUNet. To enhance the generalization ability of AttXUNet, the AttXUNet was trained on transformed dataset generated by CycleGAN for reducing the distribution difference between source domain and the target domain. We tested the proposed model on three datasets of fundus images, and the results demonstrated that our model could accurately segment DR lesions on fundus images and alleviate the degradation of segmentation performance on multiple target domains.

Citation Download Citation

Yun Zhang, Hao Zhou, Zhi Xie, and Yao He "Improving cross-domain diabetic retinopathy lesions segmentation based on CycleGAN augmentation", Proc. SPIE 12179, Second International Conference on Medical Imaging and Additive Manufacturing (ICMIAM 2022), 121791B (10 June 2022); https://doi.org/10.1117/12.2636505

ACCESS THE FULL ARTICLE

INSTITUTIONAL
Select your institution to access the SPIE Digital Library.

SELECT YOUR INSTITUTION

PERSONAL
Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.

PERSONAL SIGN IN

No SPIE Account? Create one

PURCHASE THIS CONTENT

SUBSCRIBE TO DIGITAL LIBRARY

50 downloads per 1-year subscription

Members: $195

Non-members: $335 ADD TO CART

25 downloads per 1 - year subscription

Members: $145

Non-members: $250 ADD TO CART

PURCHASE SINGLE ARTICLE

Includes PDF, HTML & Video, when available

Members: $17.00

Non-members: $21.00 ADD TO CART

PROCEEDINGS
10 PAGES

DOWNLOAD PAPER SAVE TO MY LIBRARY

GET CITATION

RIGHTS & PERMISSIONS

Get copyright permission Get copyright permission on Copyright Marketplace

KEYWORDS

Image segmentation

Data modeling

Performance modeling

Image enhancement

Computer simulations

Statistical modeling

Medical imaging

Show All Keywords

Keywords/Phrases

Search In:

Publication Years