Skip to main page content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2025 Jun 3:12:1584378.
doi: 10.3389/fmed.2025.1584378. eCollection 2025.

Development and evaluation of a deep learning system for screening real-world multiple abnormal findings based on ultra-widefield fundus images

Haodong Xiao #  1 , Lie Ju #  2   3 , Zupeng Lu #  4 , Shiguang Zhang #  5 , Yan Jiang  6 , Yuan Yang  1 , Xuerui Zhang  1 , Wenting Zhang  1 , Huanyu Liu  1 , Tingyi Liang  1 , Jianing Ren  1 , Jiawei Yin  1 , Xiaoyu Liu  7 , Tong Ma  2 , Lin Wang  2 , Wei Feng  2   3 , Kaimin Song  2 , Yuzhong Chen  2 , Zongyuan Ge  2 , Qian Shao  5 , Jie Peng  1 , Jili Chen  6 , Peiquan Zhao  1
Affiliations

Development and evaluation of a deep learning system for screening real-world multiple abnormal findings based on ultra-widefield fundus images

Haodong Xiao et al. Front Med (Lausanne). .

Abstract

Purpose: To develop and evaluate a deep learning system for screening multiple abnormal findings including hemorrhages, drusen, hard exudates, cotton wool spots and retinal breaks using ultra-widefield fundus images.

Methods: The system consisted of three modules: (I) quality assessment module, (II) artifact removal module and (III) lesion recognition module. In Module III, a heatmap was generated to highlight the lesion area. A total of 4,521 UWF images were used for the training and internal validation of the DL system. The system was evaluated in two external validation datasets consisting of 344 images and 894 images from two other hospitals. The performance of the system in these two datasets was compared with or without Module II.

Results: In both external validation datasets, the deep learning system made better performance when recognizing lesions on processed images after Module II than on original images without Module II. Module II-enhanced preprocessing improved Module III's five-lesion recognition performance by an average of 6.73% and 14.4% areas under the curves, 14.47% and 19.62% accuracy in the two external validations.

Conclusion: Our system showed reliable performance for detecting MAF in real-world UWF images. For deep learning systems to recognize real-world images, the artifact removal module was indeed helpful.

Keywords: artifact removal; deep learning; multiple abnormal findings; real-world images; ultra-widefield fundus images.

PubMed Disclaimer

Conflict of interest statement

LJ, TM, LW, WF, KS, YC, and ZG were employed by Beijing Airdoc Technology Co., Ltd. The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Figures

FIGURE 1
FIGURE 1
The overview of our proposed deep learning-based screening system.
FIGURE 2
FIGURE 2
Confusion matrix for image quality assessment.
FIGURE 3
FIGURE 3
Validation loss with early stopping.
FIGURE 4
FIGURE 4
The receiver operating characteristic (ROC) of MAF on Zhenjiang Ruikang Hospital (ZRH) external validation dataset.
FIGURE 5
FIGURE 5
Areas under the curves (AUC) performance comparison of five methods.
FIGURE 6
FIGURE 6
Representative images of original/processed UWF images and corresponding heatmaps generated using gradient-weighted class activation mapping (Grad-CAM). (a) Images and heatmaps with drusen; (b) Images and heatmaps with hemorrhage; (c) Images and heatmaps with both hemorrhage and drusen.
See this image and copyright information in PMC

Similar articles

See all similar articles

References

    1. Flaxman S, Bourne R, Resnikoff S, Ackland P, Braithwaite T, Cicinelli M, et al. Global causes of blindness and distance vision impairment 1990-2020: A systematic review and meta-analysis. Lancet Glob Health. (2017) 5:e1221–34. 10.1016/S2214-109X(17)30393-5 - DOI - PubMed
    1. Vujosevic S, Aldington S, Silva P, Hernández C, Scanlon P, Peto T, et al. Screening for diabetic retinopathy: New perspectives and challenges. Lancet Diabetes Endocrinol. (2020) 8:337–47. 10.1016/S2213-8587(19)30411-5 - DOI - PubMed
    1. Fleckenstein M, Keenan T, Guymer R, Chakravarthy U, Schmitz-Valckenberg S, Klaver C, et al. Age-related macular degeneration. Nat Rev Dis Primers. (2021) 7:1–25. 10.1038/s41572-021-00265-2 - DOI - PubMed
    1. Govers B, van Huet R, Roosing S, Keijser S, Los L, den Hollander A, et al. The genetics and disease mechanisms of rhegmatogenous retinal detachment. Prog Retin Eye Res. (2023) 97:101158. 10.1016/j.preteyeres.2022.101158 - DOI - PubMed
    1. Schmidt-Erfurth U, Sadeghipour A, Gerendas B, Waldstein S, Bogunović H. Artificial intelligence in retina. Prog Retin Eye Res. (2018) 67:1–29. 10.1016/j.preteyeres.2018.07.004 - DOI - PubMed

LinkOut - more resources